Créer des agents intelligents et sécurisés avec AlloyDB

1. Introduction

Dans cet atelier de programmation, vous allez créer un système intelligent de recherche et de recommandation de sessions de conférence à l'aide d'AlloyDB pour PostgreSQL et de ses fonctionnalités d'IA. Vous découvrirez comment combiner la recherche traditionnelle par mots clés avec la recherche vectorielle sémantique avancée, comment utiliser QueryData pour générer des instructions SQL prévisibles à partir du langage naturel et comment utiliser des fonctions d'opérateur intelligentes.

Objectifs de l'atelier

• Déployez un cluster AlloyDB et activez les fonctionnalités d'IA.
• Chargez un ensemble de données de conférence et comprenez sa structure.
• Activez l'API d'accès aux données AlloyDB.
• Utilisez les opérateurs d'IA AlloyDB, comme ai.if et ai.rank, pour les actions sémantiques.
• Implémentez la recherche hybride combinant les recherches vectorielles sémantiques (ScaNN) et textuelles (RUM).
• Activer QueryData pour AlloyDB
• Générer des modèles QueryData
• Utiliser QueryData avec des agents d'IA

Prérequis

• Un navigateur Web (par exemple, Chrome)
• Un projet Google Cloud avec facturation activée

Cet atelier de programmation s'adresse aux développeurs de tous niveaux, y compris aux débutants.

Durée totale estimée : 60 à 70 minutes. Coût estimé : moins de 3 $ (les ressources créées dans cet atelier de programmation sont éligibles à l'utilisation de l'essai ou du niveau standard sans frais).

2. Préparation

Configuration du projet

Connectez-vous à la console Google Cloud. (Si vous ne possédez pas encore de compte Gmail ou Google Workspace, vous devez en créer un.)

Utilisez un compte personnel au lieu d'un compte professionnel ou scolaire.

Attention : Il est possible que des restrictions vous empêchent d'activer les API nécessaires pour cet atelier ou d'accéder à certaines ressources avec votre compte professionnel ou scolaire.

Créer un projet Google Cloud

1. Dans la console Google Cloud, sur la page de sélection du projet, sélectionnez ou créez un projet Google Cloud.
2. Assurez-vous que la facturation est activée pour votre projet Cloud. Découvrez comment vérifier si la facturation est activée sur un projet.

Attention : Un ID de projet est unique. Une fois que vous l'avez choisi, personne d'autre ne peut l'employer. Vous en êtes le seul utilisateur. Même si un projet est supprimé, son ID ne peut pas être réutilisé.

Remarque : Si vous utilisez un compte Gmail, vous pouvez conserver l'emplacement par défaut défini sur Aucune organisation. Si vous utilisez un compte Google Workspace, choisissez un emplacement qui ait du sens pour votre organisation.

Démarrer Cloud Shell

Bien que Google Cloud puisse être utilisé à distance depuis votre ordinateur portable, nous allons nous servir de Google Cloud Shell pour cet atelier de programmation, un environnement de ligne de commande exécuté dans le cloud.

1. Cliquez sur Activer Cloud Shell en haut de la console Google Cloud.
2. Vérifiez l'authentification :

gcloud auth list

3. Confirmez votre projet :

gcloud config get project

4. Définissez-le si nécessaire :

export PROJECT_ID=<YOUR_PROJECT_ID>
gcloud config set project $PROJECT_ID

3. Avant de commencer

Activer les API

Exécutez cette commande pour activer toutes les API requises :

gcloud services enable alloydb.googleapis.com \
                       compute.googleapis.com \
                       cloudresourcemanager.googleapis.com \
                       servicenetworking.googleapis.com \
                       aiplatform.googleapis.com \
                       geminidataanalytics.googleapis.com

Présentation des API

• L'API AlloyDB (alloydb.googleapis.com) vous permet de créer, de gérer et de mettre à l'échelle des clusters AlloyDB pour PostgreSQL. Il s'agit d'un service de base de données entièrement géré et compatible avec PostgreSQL, conçu pour les charges de travail transactionnelles et analytiques d'entreprise exigeantes.
• L'API Compute Engine (compute.googleapis.com) vous permet de créer et de gérer des machines virtuelles (VM), des disques persistants et des paramètres réseau. Elle fournit la base IaaS (Infrastructure as a Service) requise pour exécuter vos charges de travail et héberger l'infrastructure sous-jacente de nombreux services gérés.
• L'API Cloud Resource Manager (cloudresourcemanager.googleapis.com) vous permet de gérer de façon programmatique les métadonnées et la configuration de votre projet Google Cloud. Il vous permet d'organiser les ressources, de gérer les stratégies IAM (Identity and Access Management) et de valider les autorisations dans la hiérarchie des projets.
• L'API Service Networking (servicenetworking.googleapis.com) vous permet d'automatiser la configuration de la connectivité privée entre votre réseau Virtual Private Cloud (VPC) et les services gérés de Google. Il est spécifiquement requis pour établir un accès par adresse IP privée pour des services tels qu'AlloyDB, afin qu'ils puissent communiquer de manière sécurisée avec vos autres ressources.
• L'API Vertex AI (aiplatform.googleapis.com) permet à vos applications de créer, de déployer et de mettre à l'échelle des modèles de machine learning. Elle fournit une interface unifiée pour tous les services d'IA de Google Cloud, y compris l'accès aux modèles d'IA générative (comme Gemini) et l'entraînement de modèles personnalisés.
• L'API Data Analytics (geminidataanalytics.googleapis.com) permet à votre application d'utiliser des fonctionnalités d'IA générales dans les produits BI.

4. Provisionner AlloyDB

Créez un cluster et une instance principale AlloyDB.

Créer une plage d'adresses IP privées

AlloyDB nécessite une plage d'adresses IP privées dans votre VPC. En supposant que vous utilisiez le réseau VPC default :

1. Créez la plage d'adresses IP privées :

gcloud compute addresses create psa-range \
    --global \
    --purpose=VPC_PEERING \
    --prefix-length=24 \
    --description="VPC private service access" \
    --network=default

2. Établissez une connexion privée :

gcloud services vpc-peerings connect \
    --service=servicenetworking.googleapis.com \
    --ranges=psa-range \
    --network=default

Créer un cluster AlloyDB

1. Créez un mot de passe pour l'utilisateur postgres :

export PGPASSWORD=`openssl rand -hex 12`
echo $PGPASSWORD

Notez ce mot de passe. Vous en aurez besoin pour vous connecter à la base de données ultérieurement.

2. Créez un cluster d'essai sans frais ("TRIAL") ou un cluster standard ("STANDARD") si ce n'est pas la première fois :

Remarque : Si vous n'avez pas encore utilisé AlloyDB, vous pouvez créer un cluster d'essai sans frais.

export REGION=us-central1
export ADBCLUSTER=alloydb-next26-ai-demo-01

gcloud alloydb clusters create $ADBCLUSTER \
    --password=$PGPASSWORD \
    --network=default \
    --region=$REGION \
    --subscription-type=TRIAL

3. Créez l'instance principale :

gcloud alloydb instances create $ADBCLUSTER-pr \
    --instance-type=PRIMARY \
    --cpu-count=8 \
    --region=$REGION \
    --cluster=$ADBCLUSTER

La création de l'instance prend entre 6 et 10 minutes.

5. Configurer les autorisations liées aux bases de données

Activer les autorisations Vertex AI pour la génération d'embeddings

PROJECT_ID=$(gcloud config get-value project)
gcloud projects add-iam-policy-binding $PROJECT_ID \
  --member="serviceAccount:service-$(gcloud projects describe $PROJECT_ID --format="value(projectNumber)")@gcp-sa-alloydb.iam.gserviceaccount.com" \
  --role="roles/aiplatform.user"

Activer l'API Data Access

Vous devez activer l'API Data Access sur le cluster AlloyDB pour pouvoir utiliser les contextes QueryData afin de créer des modèles permettant de générer des instructions SQL prévisibles à partir du langage naturel.

Dans le même onglet de terminal, exécutez :

PROJECT_ID=$(gcloud config get-value project)
REGION=us-central1
ADBCLUSTER=alloydb-next26-ai-demo-01
curl -X PATCH \
 -H "Authorization: Bearer $(gcloud auth print-access-token)" \
 -H "Content-Type: application/json" \
 https://alloydb.googleapis.com/v1alpha/projects/$PROJECT_ID/locations/$REGION/clusters/$ADBCLUSTER/instances/$ADBCLUSTER-pr?updateMask=dataApiAccess \
 -d '{
   "dataApiAccess": "ENABLED",
 }'

Activer l'authentification IAM

Pour nos outils agentiques, vous devez activer l'authentification IAM sur l'instance, puis vous ajouter en tant qu'utilisateur.

Activez IAM sur l'instance en exécutant la commande suivante dans le même onglet de terminal :

PROJECT_ID=$(gcloud config get-value project)
REGION=us-central1
ADBCLUSTER=alloydb-next26-ai-demo-01
gcloud beta alloydb instances update $ADBCLUSTER-pr \
   --database-flags alloydb.iam_authentication=on \
   --region=$REGION \
   --cluster=$ADBCLUSTER \
   --project=$PROJECT_ID \
   --update-mode=FORCE_APPLY

Ajoutez-vous en tant qu'utilisateur AlloyDB :

REGION=us-central1
ADBCLUSTER=alloydb-next26-ai-demo-01
gcloud alloydb users create $(gcloud config get-value account) \
--cluster=$ADBCLUSTER \
--region=$REGION \
--type=IAM_BASED \
--db-roles=alloydbsuperuser

6. Préparer l'exemple de base de données

Se connecter à AlloyDB Studio

1. Accédez à la page AlloyDB pour PostgreSQL de la console Google Cloud.
2. Cliquez sur votre instance principale.
3. Dans le panneau de navigation de gauche, cliquez sur AlloyDB Studio.
4. Sélectionnez la base de données postgres.
5. S'authentifier avec IAM database authentication

Créer une base de données

Exécutez le code SQL suivant dans l'éditeur de requête :

CREATE DATABASE conference_db;

Passez à la base de données conference_db :

1. Cliquez sur le bouton Current user en haut à gauche de l'écran AlloyDB studio.
2. Cliquez sur le bouton Switch user/database.
3. Sélectionnez la base de données conference_db que vous venez de créer.

Activer pgvector

Assurez-vous que l'extension vector standard est activée :

CREATE EXTENSION IF NOT EXISTS vector;

Charger des exemples de données

Exécutez les scripts SQL suivants pour créer le schéma et le remplir avec des exemples de données :

1. Nettoyer les tables conflictuelles précédentes

DROP TABLE IF EXISTS public.attendees_sessions CASCADE;
DROP TABLE IF EXISTS public.session_speaker_mapping CASCADE;
DROP TABLE IF EXISTS public.session_topic_mapping CASCADE;
DROP TABLE IF EXISTS public.session CASCADE;
DROP TABLE IF EXISTS public.attendees CASCADE;
DROP TABLE IF EXISTS public.speaker CASCADE;
DROP TABLE IF EXISTS public.session_topic CASCADE;

2. Créer des tables

CREATE TABLE public.attendees (
    username character varying(100) NOT NULL PRIMARY KEY,
    name character varying(100),
    company character varying(150),
    job_title character varying(100),
    area_of_interest character varying(100),
    street_address text,
    city character varying(100),
    state_province character varying(100),
    country character varying(100),
    created_at timestamp without time zone DEFAULT CURRENT_TIMESTAMP
);

CREATE TABLE public.attendees_sessions (
    username character varying(100) NOT NULL REFERENCES public.attendees(username),
    session_id character varying(50) NOT NULL,
    registration_date timestamp without time zone DEFAULT CURRENT_TIMESTAMP,
    PRIMARY KEY (username, session_id)
);

CREATE TABLE public.session (
    session_id character varying(50) NOT NULL PRIMARY KEY,
    session_name character varying(255) NOT NULL,
    full_description text,
    full_description_embedding vector(768) GENERATED ALWAYS AS (embedding('text-embedding-005', full_description)) STORED,
    description_tsvector tsvector GENERATED ALWAYS AS (to_tsvector('english', full_description)) STORED,
    session_format character varying(50),
    learning_level character varying(50),
    session_url character varying(255),
    for_job_role character varying(250),
    session_date date,
    session_start_time time without time zone,
    session_end_time time without time zone,
    session_location character varying(100),
    capacity integer,
    remaining_capacity integer,
    interest_area character varying(100),
    industry character varying(100)
);

COMMENT ON COLUMN public.session.session_format IS 'The format of a session. Possible values are Keynotes, Breakouts, Lightning Talks.';

CREATE TABLE public.speaker (
    speaker_id integer NOT NULL PRIMARY KEY,
    speaker_name character varying(100) NOT NULL,
    job_title character varying(100),
    company character varying(100),
    type character varying(50),
    profile_pic character varying(100),
    speaker_url character varying(100)
);

CREATE TABLE public.session_speaker_mapping (
    session_id character varying(50) NOT NULL REFERENCES public.session(session_id),
    speaker_id integer NOT NULL REFERENCES public.speaker(speaker_id),
    PRIMARY KEY (session_id, speaker_id)
);

CREATE TABLE public.session_topic (
    topic character varying(100) NOT NULL PRIMARY KEY,
    topic_desc text,
    embedding vector(768) GENERATED ALWAYS AS (embedding('text-embedding-005', topic_desc)) STORED
);

CREATE TABLE public.session_topic_mapping (
    session_id character varying(50) NOT NULL REFERENCES public.session(session_id),
    topic character varying(100) NOT NULL REFERENCES public.session_topic(topic),
    PRIMARY KEY (session_id, topic)
);

3. Remplir les tables avec des exemples de données

3.1 Remplir les thèmes des sessions

-- Insert Topics
INSERT INTO public.session_topic (topic, topic_desc) VALUES
('Databases', 'Relational and non-relational database technologies.'),
('AI & Machine Learning', 'Generative AI, LLMs, and ML infrastructure.'),
('Cloud Architecture', 'Designing scalable and resilient cloud systems.'),
('Security', 'Identity, compliance, and network security.'),
('DevOps', 'CI/CD, platform engineering, and automation.') ON CONFLICT DO NOTHING;

3.2 Renseigner les enceintes

-- Insert Speakers
INSERT INTO public.speaker (speaker_id, speaker_name, job_title, company, type, profile_pic, speaker_url) VALUES
(1, 'Speaker 1', 'Director of Engineering', 'DeepMind', 'External', 'pic_1.png', 'http://speakers.com/1'),
(2, 'Speaker 2', 'Cloud Architect', 'Verily', 'External', 'pic_2.png', 'http://speakers.com/2'),
(3, 'Speaker 3', 'Product Manager', 'GlobalEnterprises', 'External', 'pic_3.png', 'http://speakers.com/3'),
(4, 'Speaker 4', 'Tech Lead', 'Google', 'Internal', 'pic_4.png', 'http://speakers.com/4'),
(5, 'Speaker 5', 'Security Engineer', 'DeepMind', 'External', 'pic_5.png', 'http://speakers.com/5'),
(6, 'Speaker 6', 'DevOps Engineer', 'DeepMind', 'External', 'pic_6.png', 'http://speakers.com/6'),
(7, 'Speaker 7', 'Product Manager', 'DataSystems', 'External', 'pic_7.png', 'http://speakers.com/7'),
(8, 'Speaker 8', 'Director of Engineering', 'Google', 'Internal', 'pic_8.png', 'http://speakers.com/8'),
(9, 'Speaker 9', 'Cloud Architect', 'Google', 'Internal', 'pic_9.png', 'http://speakers.com/9'),
(10, 'Speaker 10', 'DevOps Engineer', 'Alphabet', 'Internal', 'pic_10.png', 'http://speakers.com/10'),
(11, 'Speaker 11', 'Tech Lead', 'Alphabet', 'External', 'pic_11.png', 'http://speakers.com/11'),
(12, 'Speaker 12', 'Product Manager', 'Alphabet', 'External', 'pic_12.png', 'http://speakers.com/12'),
(13, 'Speaker 13', 'Tech Lead', 'SoftSolutions', 'Internal', 'pic_13.png', 'http://speakers.com/13'),
(14, 'Speaker 14', 'Director of Engineering', 'Verily', 'Internal', 'pic_14.png', 'http://speakers.com/14'),
(15, 'Speaker 15', 'DevOps Engineer', 'CloudNative', 'External', 'pic_15.png', 'http://speakers.com/15'),
(16, 'Speaker 16', 'Software Engineer', 'Waymo', 'External', 'pic_16.png', 'http://speakers.com/16'),
(17, 'Speaker 17', 'DevOps Engineer', 'Google', 'External', 'pic_17.png', 'http://speakers.com/17'),
(18, 'Speaker 18', 'Data Scientist', 'GlobalEnterprises', 'External', 'pic_18.png', 'http://speakers.com/18'),
(19, 'Speaker 19', 'Security Engineer', 'Waymo', 'Internal', 'pic_19.png', 'http://speakers.com/19'),
(20, 'Speaker 20', 'Tech Lead', 'SoftSolutions', 'Internal', 'pic_20.png', 'http://speakers.com/20') ON CONFLICT (speaker_id) DO NOTHING;

3.3 Ajouter des participants

-- Insert Attendees
INSERT INTO public.attendees (username, name, company, job_title, area_of_interest, street_address, city, state_province, country) VALUES
('user_1', 'Attendee 1', 'GlobalEnterprises', 'Data Scientist', 'DevOps', '1 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_2', 'Attendee 2', 'Alphabet', 'Director of Engineering', 'DevOps', '2 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_3', 'Attendee 3', 'CloudNative', 'Data Scientist', 'AI & Machine Learning', '3 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_4', 'Attendee 4', 'SoftSolutions', 'Product Manager', 'DevOps', '4 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_5', 'Attendee 5', 'DataSystems', 'Software Engineer', 'DevOps', '5 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_6', 'Attendee 6', 'Google', 'Director of Engineering', 'Security', '6 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_7', 'Attendee 7', 'Waymo', 'Product Manager', 'AI & Machine Learning', '7 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_8', 'Attendee 8', 'Verily', 'Product Manager', 'DevOps', '8 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_9', 'Attendee 9', 'Waymo', 'Software Engineer', 'DevOps', '9 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_10', 'Attendee 10', 'Waymo', 'DevOps Engineer', 'Databases', '10 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_11', 'Attendee 11', 'GlobalEnterprises', 'Tech Lead', 'Cloud Architecture', '11 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_12', 'Attendee 12', 'Google', 'Tech Lead', 'DevOps', '12 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_13', 'Attendee 13', 'DeepMind', 'Software Engineer', 'Cloud Architecture', '13 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_14', 'Attendee 14', 'DataSystems', 'Security Engineer', 'DevOps', '14 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_15', 'Attendee 15', 'DeepMind', 'Security Engineer', 'Cloud Architecture', '15 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_16', 'Attendee 16', 'Verily', 'Security Engineer', 'Cloud Architecture', '16 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_17', 'Attendee 17', 'Google', 'Software Engineer', 'Cloud Architecture', '17 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_18', 'Attendee 18', 'GlobalEnterprises', 'Security Engineer', 'DevOps', '18 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_19', 'Attendee 19', 'CloudNative', 'Data Scientist', 'Databases', '19 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_20', 'Attendee 20', 'DataSystems', 'Cloud Architect', 'AI & Machine Learning', '20 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_21', 'Attendee 21', 'Google', 'Data Scientist', 'Databases', '21 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_22', 'Attendee 22', 'TechCorp', 'Data Scientist', 'DevOps', '22 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_23', 'Attendee 23', 'DeepMind', 'Product Manager', 'Databases', '23 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_24', 'Attendee 24', 'Google', 'Cloud Architect', 'AI & Machine Learning', '24 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_25', 'Attendee 25', 'GlobalEnterprises', 'Software Engineer', 'AI & Machine Learning', '25 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_26', 'Attendee 26', 'Verily', 'Security Engineer', 'Security', '26 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_27', 'Attendee 27', 'GlobalEnterprises', 'Tech Lead', 'Databases', '27 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_28', 'Attendee 28', 'Google', 'Tech Lead', 'Databases', '28 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_29', 'Attendee 29', 'SoftSolutions', 'Software Engineer', 'Databases', '29 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_30', 'Attendee 30', 'Verily', 'DevOps Engineer', 'Databases', '30 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_31', 'Attendee 31', 'DeepMind', 'Data Scientist', 'Databases', '31 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_32', 'Attendee 32', 'SoftSolutions', 'Software Engineer', 'Cloud Architecture', '32 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_33', 'Attendee 33', 'DeepMind', 'Tech Lead', 'Databases', '33 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_34', 'Attendee 34', 'Alphabet', 'Security Engineer', 'DevOps', '34 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_35', 'Attendee 35', 'TechCorp', 'DevOps Engineer', 'Cloud Architecture', '35 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_36', 'Attendee 36', 'DataSystems', 'Director of Engineering', 'Security', '36 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_37', 'Attendee 37', 'Google', 'Cloud Architect', 'Databases', '37 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_38', 'Attendee 38', 'Google', 'Product Manager', 'Security', '38 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_39', 'Attendee 39', 'SoftSolutions', 'Security Engineer', 'Cloud Architecture', '39 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_40', 'Attendee 40', 'Waymo', 'Director of Engineering', 'Cloud Architecture', '40 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_41', 'Attendee 41', 'SoftSolutions', 'Tech Lead', 'Security', '41 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_42', 'Attendee 42', 'DeepMind', 'Cloud Architect', 'Security', '42 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_43', 'Attendee 43', 'TechCorp', 'Cloud Architect', 'Databases', '43 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_44', 'Attendee 44', 'Alphabet', 'Security Engineer', 'AI & Machine Learning', '44 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_45', 'Attendee 45', 'CloudNative', 'Cloud Architect', 'AI & Machine Learning', '45 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_46', 'Attendee 46', 'CloudNative', 'Data Scientist', 'Security', '46 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_47', 'Attendee 47', 'DeepMind', 'Data Scientist', 'Security', '47 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_48', 'Attendee 48', 'Verily', 'DevOps Engineer', 'AI & Machine Learning', '48 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_49', 'Attendee 49', 'Verily', 'Director of Engineering', 'AI & Machine Learning', '49 Main St', 'CityVille', 'StateName', 'CountryName'),
('user_50', 'Attendee 50', 'Google', 'Director of Engineering', 'DevOps', '50 Main St', 'CityVille', 'StateName', 'CountryName') ON CONFLICT (username) DO NOTHING;

3.4 Remplir les sessions

-- Insert Sessions
INSERT INTO public.session (session_id, session_name, full_description, session_format, learning_level, session_url, for_job_role, session_date, session_start_time, session_end_time, session_location, capacity, remaining_capacity, interest_area, industry) VALUES
('S001', 'AlloyDB Deep Dive: Advanced Performance Tuning', 'Learn how to squeeze every drop of performance out of AlloyDB. This session covers index tuning, memory management, and advanced query optimization techniques.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S001', 'Cloud Architect', '2026-04-11', '15:00:00', '15:00:00', 'Room C', 500, 192, 'Databases', 'Technology'),
('S002', 'Vector Search at Scale with ScaNN in AlloyDB', 'Discover how to use ScaNN for fast approximate nearest neighbor search in AlloyDB. Perfect for building high-scale recommendation systems.', 'Breakouts', 'Intermediate', 'http://sessions.com/S002', 'Cloud Architect', '2026-04-10', '14:00:00', '14:00:00', 'Auditorium', 100, 16, 'Databases', 'Retail'),
('S003', 'Building Gen AI Apps with Vertex AI and AlloyDB', 'A practical guide to integrating Vertex AI embeddings and LLMs with your operational data in AlloyDB to build intelligent applications.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S003', 'Director of Engineering', '2026-04-11', '11:00:00', '11:00:00', 'Auditorium', 50, 28, 'Databases', 'Healthcare'),
('S004', 'Spanner: Architecting for Global Consistency', 'Learn how Spanner achieves global scale without sacrificing strong consistency. We will cover multi-region deployment patterns.', 'Keynotes', 'Intermediate', 'http://sessions.com/S004', 'Cloud Architect', '2026-04-11', '11:00:00', '11:00:00', 'Room B', 500, 441, 'Databases', 'Healthcare'),
('S005', 'BigQuery + Vertex AI: Predictive Analytics Made Easy', 'See how to combine the power of BigQuery for data warehousing with Vertex AI for machine learning to build predictive models directly on your data.', 'Breakouts', 'Advanced', 'http://sessions.com/S005', 'Product Manager', '2026-04-11', '14:00:00', '14:00:00', 'Room C', 100, 96, 'AI & Machine Learning', 'Retail'),
('S006', 'Securing Your Data: Best Practices in AlloyDB', 'Deep dive into the security features of AlloyDB, including IAM integration, encryption at rest and in transit, and audit logging.', 'Keynotes', 'Beginner', 'http://sessions.com/S006', 'Product Manager', '2026-04-09', '15:00:00', '15:00:00', 'Room C', 200, 199, 'Databases', 'Healthcare'),
('S007', 'Kubernetes for Databases: Running PostgreSQL on GKE', 'Best practices for running stateful workloads like PostgreSQL on Google Kubernetes Engine (GKE).', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S007', 'Software Engineer', '2026-04-09', '16:00:00', '16:00:00', 'Room C', 100, 63, 'Databases', 'Retail'),
('S008', 'Real-time Analytics with BigQuery and Pub/Sub', 'Learn how to build real-time data pipelines using Pub/Sub and stream data directly into BigQuery for instant insights.', 'Keynotes', 'Beginner', 'http://sessions.com/S008', 'Security Engineer', '2026-04-10', '14:00:00', '14:00:00', 'Room C', 200, 65, 'Databases', 'Finance'),
('S009', 'Microservices Architecture with Spanner', 'How to design microservices that leverage Spanner''s unique capabilities for distributed transactions and scalability.', 'Keynotes', 'Beginner', 'http://sessions.com/S009', 'Software Engineer', '2026-04-11', '10:00:00', '10:00:00', 'Auditorium', 100, 25, 'Databases', 'Technology'),
('S010', 'AI-Powered Search with ScaNN and LLMs', 'Learn how to combine ScaNN vector search with Large Language Models to create powerful, context-aware search experiences.', 'Keynotes', 'Intermediate', 'http://sessions.com/S010', 'Tech Lead', '2026-04-11', '11:00:00', '11:00:00', 'Room C', 50, 27, 'AI & Machine Learning', 'Healthcare'),
('S011', 'AlloyDB Omni: Run AlloyDB Anywhere', 'Explore AlloyDB Omni, the downloadable edition of AlloyDB that lets you run the same high-performance database in your own data center or on the edge.', 'Breakouts', 'Intermediate', 'http://sessions.com/S011', 'Cloud Architect', '2026-04-09', '10:00:00', '10:00:00', 'Room B', 200, 195, 'Databases', 'Technology'),
('S012', 'Data Mesh on Google Cloud: Best Practices', 'How to implement a decentralized data mesh architecture using BigQuery, Dataplex, and other Google Cloud tools.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S012', 'Data Scientist', '2026-04-10', '16:00:00', '16:00:00', 'Room C', 100, 46, 'Cloud Architecture', 'Technology'),
('S013', 'Serverless Databases: When to use Cloud SQL vs AlloyDB', 'A comparison of Cloud SQL and AlloyDB, helping you choose the right database for your serverless and traditional applications.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S013', 'DevOps Engineer', '2026-04-10', '11:00:00', '11:00:00', 'Room C', 500, 458, 'Databases', 'Manufacturing'),
('S014', 'Graph Databases on Google Cloud', 'Explore options for graph data processing on Google Cloud, including integrations with existing database services.', 'Breakouts', 'Advanced', 'http://sessions.com/S014', 'DevOps Engineer', '2026-04-11', '09:00:00', '09:00:00', 'Room C', 500, 338, 'DevOps', 'Finance'),
('S015', 'Automating DB Ops with Gemini', 'See how Gemini can help DBA and developers write better SQL, optimize queries, and manage database infrastructure.', 'Breakouts', 'Advanced', 'http://sessions.com/S015', 'Cloud Architect', '2026-04-11', '14:00:00', '14:00:00', 'Room C', 100, 53, 'AI & Machine Learning', 'Retail'),
('S016', 'High Availability and Disaster Recovery in AlloyDB', 'A deep dive into how AlloyDB ensures your data is always available, covering failover mechanisms and backup strategies.', 'Breakouts', 'Advanced', 'http://sessions.com/S016', 'Cloud Architect', '2026-04-11', '11:00:00', '11:00:00', 'Room A', 200, 52, 'Databases', 'Technology'),
('S017', 'Optimizing Costs in BigQuery', 'Practical tips for reducing your BigQuery bill without sacrificing performance, covering slot management and query optimization.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S017', 'Security Engineer', '2026-04-11', '15:00:00', '15:00:00', 'Room B', 100, 65, 'Databases', 'Manufacturing'),
('S018', 'Continuous Integration for Database Schemas', 'How to apply CI/CD principles to database schema changes using tools like Liquibase or Flyway on Google Cloud.', 'Breakouts', 'Beginner', 'http://sessions.com/S018', 'Cloud Architect', '2026-04-11', '09:00:00', '09:00:00', 'Room B', 50, 30, 'Security', 'Technology'),
('S019', 'Data Governance in the Age of AI', 'Learn how to maintain data quality, privacy, and compliance when feeding enterprise data into AI models.', 'Breakouts', 'Beginner', 'http://sessions.com/S019', 'Product Manager', '2026-04-10', '10:00:00', '10:00:00', 'Room A', 100, 39, 'DevOps', 'Retail'),
('S020', 'Hybrid Search: Combining Vector and Keyword Search', 'Learn how to implement hybrid search in AlloyDB to get the best of both worlds: semantic understanding and precise keyword matching.', 'Breakouts', 'Intermediate', 'http://sessions.com/S020', 'Data Scientist', '2026-04-09', '14:00:00', '14:00:00', 'Auditorium', 500, 176, 'Databases', 'Manufacturing'),
('S021', 'Deep Dive into ScaNN for High Availability', 'Join this session to explore how ScaNN can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S021', 'Software Engineer', '2026-04-11', '14:00:00', '14:00:00', 'Room B', 200, 18, 'AI & Machine Learning', 'Finance'),
('S022', 'Understanding Vertex AI for Modernization', 'Join this session to explore how Vertex AI can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S022', 'Tech Lead', '2026-04-09', '09:00:00', '09:00:00', 'Room A', 50, 48, 'AI & Machine Learning', 'Finance'),
('S023', 'Securing Vertex AI for Modernization', 'Join this session to explore how Vertex AI can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S023', 'Tech Lead', '2026-04-10', '10:00:00', '10:00:00', 'Auditorium', 200, 96, 'AI & Machine Learning', 'Manufacturing'),
('S024', 'Mastering Kubernetes for Security', 'Join this session to explore how Kubernetes can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S024', 'DevOps Engineer', '2026-04-09', '09:00:00', '09:00:00', 'Auditorium', 200, 160, 'Security', 'Technology'),
('S025', 'Mastering AlloyDB for Modernization', 'Join this session to explore how AlloyDB can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S025', 'Director of Engineering', '2026-04-11', '16:00:00', '16:00:00', 'Room B', 200, 74, 'Databases', 'Healthcare'),
('S026', 'Securing Vertex AI for Automation', 'Join this session to explore how Vertex AI can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S026', 'Cloud Architect', '2026-04-11', '16:00:00', '16:00:00', 'Room A', 50, 25, 'AI & Machine Learning', 'Manufacturing'),
('S027', 'Mastering BigQuery for Analytics', 'Join this session to explore how BigQuery can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S027', 'Security Engineer', '2026-04-11', '10:00:00', '10:00:00', 'Room B', 200, 117, 'Databases', 'Healthcare'),
('S028', 'Deep Dive into Spanner for Integration', 'Join this session to explore how Spanner can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S028', 'Tech Lead', '2026-04-09', '10:00:00', '10:00:00', 'Auditorium', 50, 32, 'Databases', 'Retail'),
('S029', 'Scaling Generative AI for Automation', 'Join this session to explore how Generative AI can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S029', 'Tech Lead', '2026-04-09', '10:00:00', '10:00:00', 'Room A', 200, 157, 'AI & Machine Learning', 'Healthcare'),
('S030', 'Deploying Generative AI for Analytics', 'Join this session to explore how Generative AI can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S030', 'Director of Engineering', '2026-04-09', '16:00:00', '16:00:00', 'Room B', 500, 462, 'DevOps', 'Healthcare'),
('S031', 'Mastering ScaNN for High Availability', 'Join this session to explore how ScaNN can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S031', 'Director of Engineering', '2026-04-10', '09:00:00', '09:00:00', 'Room C', 500, 140, 'AI & Machine Learning', 'Finance'),
('S032', 'Deploying Kubernetes for Integration', 'Join this session to explore how Kubernetes can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S032', 'Cloud Architect', '2026-04-11', '14:00:00', '14:00:00', 'Auditorium', 200, 22, 'Databases', 'Retail'),
('S033', 'Securing Gemini for Performance', 'Join this session to explore how Gemini can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S033', 'Cloud Architect', '2026-04-09', '16:00:00', '16:00:00', 'Room B', 200, 84, 'AI & Machine Learning', 'Technology'),
('S034', 'Deploying Vertex AI for Scale', 'Join this session to explore how Vertex AI can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S034', 'Director of Engineering', '2026-04-10', '10:00:00', '10:00:00', 'Room B', 500, 222, 'AI & Machine Learning', 'Finance'),
('S035', 'Deep Dive into BigQuery for Analytics', 'Join this session to explore how BigQuery can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S035', 'Security Engineer', '2026-04-11', '15:00:00', '15:00:00', 'Room A', 500, 408, 'Databases', 'Healthcare'),
('S036', 'Deploying Kubernetes for Security', 'Join this session to explore how Kubernetes can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S036', 'Tech Lead', '2026-04-10', '09:00:00', '09:00:00', 'Room A', 200, 54, 'Security', 'Healthcare'),
('S037', 'Understanding Spanner for Integration', 'Join this session to explore how Spanner can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S037', 'Tech Lead', '2026-04-10', '14:00:00', '14:00:00', 'Room C', 200, 89, 'Databases', 'Technology'),
('S038', 'Deep Dive into Kubernetes for Automation', 'Join this session to explore how Kubernetes can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S038', 'Product Manager', '2026-04-11', '09:00:00', '09:00:00', 'Room C', 50, 47, 'AI & Machine Learning', 'Healthcare'),
('S039', 'Optimizing BigQuery for Reliability', 'Join this session to explore how BigQuery can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S039', 'Director of Engineering', '2026-04-10', '14:00:00', '14:00:00', 'Auditorium', 200, 38, 'Databases', 'Retail'),
('S040', 'Deep Dive into Spanner for Cost Efficiency', 'Join this session to explore how Spanner can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S040', 'Software Engineer', '2026-04-09', '14:00:00', '14:00:00', 'Auditorium', 50, 2, 'Databases', 'Healthcare'),
('S041', 'Exploring Gemini for Cost Efficiency', 'Join this session to explore how Gemini can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S041', 'Product Manager', '2026-04-10', '15:00:00', '15:00:00', 'Room A', 100, 73, 'AI & Machine Learning', 'Healthcare'),
('S042', 'Mastering ScaNN for Performance', 'Join this session to explore how ScaNN can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S042', 'Tech Lead', '2026-04-09', '11:00:00', '11:00:00', 'Auditorium', 100, 50, 'AI & Machine Learning', 'Manufacturing'),
('S043', 'Mastering ScaNN for Analytics', 'Join this session to explore how ScaNN can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S043', 'Data Scientist', '2026-04-10', '09:00:00', '09:00:00', 'Room B', 200, 111, 'AI & Machine Learning', 'Technology'),
('S044', 'Securing Gemini for Integration', 'Join this session to explore how Gemini can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S044', 'Cloud Architect', '2026-04-09', '14:00:00', '14:00:00', 'Auditorium', 500, 281, 'AI & Machine Learning', 'Healthcare'),
('S045', 'Architecting Cloud SQL for Integration', 'Join this session to explore how Cloud SQL can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S045', 'Product Manager', '2026-04-11', '11:00:00', '11:00:00', 'Room C', 50, 6, 'AI & Machine Learning', 'Retail'),
('S046', 'Securing Kubernetes for Scale', 'Join this session to explore how Kubernetes can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S046', 'Security Engineer', '2026-04-11', '14:00:00', '14:00:00', 'Room B', 100, 80, 'AI & Machine Learning', 'Manufacturing'),
('S047', 'Mastering PostgreSQL for Integration', 'Join this session to explore how PostgreSQL can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S047', 'Tech Lead', '2026-04-10', '14:00:00', '14:00:00', 'Auditorium', 200, 134, 'Databases', 'Manufacturing'),
('S048', 'Deep Dive into Kubernetes for Cost Efficiency', 'Join this session to explore how Kubernetes can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S048', 'Data Scientist', '2026-04-11', '09:00:00', '09:00:00', 'Room C', 500, 102, 'AI & Machine Learning', 'Retail'),
('S049', 'Deep Dive into Cloud SQL for Scale', 'Join this session to explore how Cloud SQL can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S049', 'Security Engineer', '2026-04-09', '14:00:00', '14:00:00', 'Auditorium', 200, 127, 'AI & Machine Learning', 'Healthcare'),
('S050', 'Scaling Cloud SQL for Reliability', 'Join this session to explore how Cloud SQL can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S050', 'Software Engineer', '2026-04-10', '15:00:00', '15:00:00', 'Room A', 100, 40, 'Cloud Architecture', 'Finance'),
('S051', 'Building Vertex AI for Security', 'Join this session to explore how Vertex AI can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S051', 'Director of Engineering', '2026-04-09', '15:00:00', '15:00:00', 'Room C', 500, 138, 'AI & Machine Learning', 'Retail'),
('S052', 'Exploring Vertex AI for Modernization', 'Join this session to explore how Vertex AI can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S052', 'DevOps Engineer', '2026-04-09', '09:00:00', '09:00:00', 'Room C', 200, 27, 'AI & Machine Learning', 'Finance'),
('S053', 'Optimizing Generative AI for Automation', 'Join this session to explore how Generative AI can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S053', 'Director of Engineering', '2026-04-10', '11:00:00', '11:00:00', 'Room C', 500, 65, 'Databases', 'Healthcare'),
('S054', 'Architecting PostgreSQL for Performance', 'Join this session to explore how PostgreSQL can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S054', 'Data Scientist', '2026-04-09', '11:00:00', '11:00:00', 'Auditorium', 200, 103, 'Databases', 'Healthcare'),
('S055', 'Mastering Generative AI for Automation', 'Join this session to explore how Generative AI can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S055', 'Data Scientist', '2026-04-11', '11:00:00', '11:00:00', 'Room C', 50, 31, 'AI & Machine Learning', 'Technology'),
('S056', 'Exploring ScaNN for Reliability', 'Join this session to explore how ScaNN can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S056', 'Security Engineer', '2026-04-11', '15:00:00', '15:00:00', 'Room A', 500, 351, 'AI & Machine Learning', 'Healthcare'),
('S057', 'Deep Dive into BigQuery for Performance', 'Join this session to explore how BigQuery can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S057', 'DevOps Engineer', '2026-04-11', '11:00:00', '11:00:00', 'Room B', 200, 200, 'Databases', 'Finance'),
('S058', 'Scaling Generative AI for Reliability', 'Join this session to explore how Generative AI can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S058', 'Cloud Architect', '2026-04-10', '11:00:00', '11:00:00', 'Room C', 100, 61, 'Databases', 'Finance'),
('S059', 'Securing Vertex AI for High Availability', 'Join this session to explore how Vertex AI can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S059', 'Data Scientist', '2026-04-11', '16:00:00', '16:00:00', 'Room B', 500, 224, 'AI & Machine Learning', 'Manufacturing'),
('S060', 'Architecting AlloyDB for Integration', 'Join this session to explore how AlloyDB can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S060', 'Tech Lead', '2026-04-09', '14:00:00', '14:00:00', 'Room A', 500, 223, 'Databases', 'Technology'),
('S061', 'Understanding Spanner for High Availability', 'Join this session to explore how Spanner can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S061', 'Tech Lead', '2026-04-09', '16:00:00', '16:00:00', 'Room C', 100, 52, 'Databases', 'Manufacturing'),
('S062', 'Scaling Kubernetes for Cost Efficiency', 'Join this session to explore how Kubernetes can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S062', 'Product Manager', '2026-04-10', '16:00:00', '16:00:00', 'Room B', 100, 0, 'Cloud Architecture', 'Retail'),
('S063', 'Architecting Kubernetes for High Availability', 'Join this session to explore how Kubernetes can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S063', 'Security Engineer', '2026-04-09', '10:00:00', '10:00:00', 'Auditorium', 200, 0, 'Security', 'Finance'),
('S064', 'Mastering Gemini for Integration', 'Join this session to explore how Gemini can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S064', 'Security Engineer', '2026-04-10', '15:00:00', '15:00:00', 'Room B', 100, 8, 'AI & Machine Learning', 'Technology'),
('S065', 'Optimizing Generative AI for Security', 'Join this session to explore how Generative AI can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S065', 'Tech Lead', '2026-04-09', '15:00:00', '15:00:00', 'Room C', 500, 230, 'AI & Machine Learning', 'Retail'),
('S066', 'Optimizing BigQuery for Scale', 'Join this session to explore how BigQuery can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S066', 'Tech Lead', '2026-04-11', '10:00:00', '10:00:00', 'Room C', 500, 186, 'Databases', 'Healthcare'),
('S067', 'Scaling PostgreSQL for Automation', 'Join this session to explore how PostgreSQL can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S067', 'Product Manager', '2026-04-10', '11:00:00', '11:00:00', 'Auditorium', 100, 17, 'Databases', 'Retail'),
('S068', 'Exploring AlloyDB for Automation', 'Join this session to explore how AlloyDB can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S068', 'Director of Engineering', '2026-04-11', '16:00:00', '16:00:00', 'Room B', 100, 77, 'Databases', 'Retail'),
('S069', 'Securing BigQuery for Cost Efficiency', 'Join this session to explore how BigQuery can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S069', 'Cloud Architect', '2026-04-09', '14:00:00', '14:00:00', 'Room A', 500, 437, 'Databases', 'Manufacturing'),
('S070', 'Deep Dive into Kubernetes for Integration', 'Join this session to explore how Kubernetes can be used for Integration. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S070', 'Cloud Architect', '2026-04-11', '11:00:00', '11:00:00', 'Room A', 50, 46, 'AI & Machine Learning', 'Technology'),
('S071', 'Architecting ScaNN for Cost Efficiency', 'Join this session to explore how ScaNN can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S071', 'Cloud Architect', '2026-04-11', '14:00:00', '14:00:00', 'Auditorium', 100, 97, 'AI & Machine Learning', 'Finance'),
('S072', 'Scaling ScaNN for Analytics', 'Join this session to explore how ScaNN can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S072', 'Cloud Architect', '2026-04-09', '11:00:00', '11:00:00', 'Room C', 100, 96, 'AI & Machine Learning', 'Manufacturing'),
('S073', 'Securing Cloud SQL for Modernization', 'Join this session to explore how Cloud SQL can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S073', 'Tech Lead', '2026-04-10', '16:00:00', '16:00:00', 'Room C', 200, 195, 'Security', 'Technology'),
('S074', 'Deploying ScaNN for Scale', 'Join this session to explore how ScaNN can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S074', 'Data Scientist', '2026-04-10', '10:00:00', '10:00:00', 'Room A', 500, 215, 'AI & Machine Learning', 'Technology'),
('S075', 'Architecting BigQuery for Automation', 'Join this session to explore how BigQuery can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S075', 'Product Manager', '2026-04-09', '09:00:00', '09:00:00', 'Room A', 200, 110, 'Databases', 'Healthcare'),
('S076', 'Understanding Vertex AI for Scale', 'Join this session to explore how Vertex AI can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S076', 'Software Engineer', '2026-04-11', '15:00:00', '15:00:00', 'Room B', 200, 15, 'AI & Machine Learning', 'Finance'),
('S077', 'Architecting PostgreSQL for Reliability', 'Join this session to explore how PostgreSQL can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S077', 'Director of Engineering', '2026-04-09', '11:00:00', '11:00:00', 'Room C', 200, 133, 'Databases', 'Finance'),
('S078', 'Deploying Generative AI for Analytics', 'Join this session to explore how Generative AI can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S078', 'Product Manager', '2026-04-11', '16:00:00', '16:00:00', 'Room A', 200, 190, 'AI & Machine Learning', 'Manufacturing'),
('S079', 'Deploying ScaNN for Analytics', 'Join this session to explore how ScaNN can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S079', 'Cloud Architect', '2026-04-11', '16:00:00', '16:00:00', 'Room B', 200, 95, 'AI & Machine Learning', 'Technology'),
('S080', 'Securing Spanner for Security', 'Join this session to explore how Spanner can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S080', 'Tech Lead', '2026-04-11', '14:00:00', '14:00:00', 'Room A', 500, 310, 'Databases', 'Healthcare'),
('S081', 'Optimizing Cloud SQL for Performance', 'Join this session to explore how Cloud SQL can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S081', 'Security Engineer', '2026-04-11', '11:00:00', '11:00:00', 'Room B', 100, 76, 'Databases', 'Finance'),
('S082', 'Exploring Spanner for Cost Efficiency', 'Join this session to explore how Spanner can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S082', 'Tech Lead', '2026-04-11', '15:00:00', '15:00:00', 'Room C', 50, 1, 'Databases', 'Manufacturing'),
('S083', 'Understanding Kubernetes for Performance', 'Join this session to explore how Kubernetes can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S083', 'Product Manager', '2026-04-11', '14:00:00', '14:00:00', 'Auditorium', 100, 49, 'Security', 'Retail'),
('S084', 'Exploring Cloud SQL for Cost Efficiency', 'Join this session to explore how Cloud SQL can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S084', 'Tech Lead', '2026-04-09', '16:00:00', '16:00:00', 'Room C', 500, 91, 'Cloud Architecture', 'Healthcare'),
('S085', 'Scaling Kubernetes for Automation', 'Join this session to explore how Kubernetes can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Intermediate', 'http://sessions.com/S085', 'Director of Engineering', '2026-04-10', '10:00:00', '10:00:00', 'Room B', 500, 204, 'Databases', 'Finance'),
('S086', 'Deep Dive into BigQuery for Cost Efficiency', 'Join this session to explore how BigQuery can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S086', 'Product Manager', '2026-04-09', '16:00:00', '16:00:00', 'Room B', 100, 89, 'Databases', 'Finance'),
('S087', 'Deep Dive into Vertex AI for Automation', 'Join this session to explore how Vertex AI can be used for Automation. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S087', 'Data Scientist', '2026-04-10', '14:00:00', '14:00:00', 'Auditorium', 100, 74, 'AI & Machine Learning', 'Retail'),
('S088', 'Mastering AlloyDB for Modernization', 'Join this session to explore how AlloyDB can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S088', 'Security Engineer', '2026-04-09', '16:00:00', '16:00:00', 'Room C', 100, 39, 'Databases', 'Healthcare'),
('S089', 'Exploring AlloyDB for High Availability', 'Join this session to explore how AlloyDB can be used for High Availability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S089', 'Data Scientist', '2026-04-10', '15:00:00', '15:00:00', 'Room A', 100, 94, 'Databases', 'Healthcare'),
('S090', 'Exploring Vertex AI for Security', 'Join this session to explore how Vertex AI can be used for Security. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Intermediate', 'http://sessions.com/S090', 'Security Engineer', '2026-04-11', '10:00:00', '10:00:00', 'Room B', 50, 27, 'AI & Machine Learning', 'Retail'),
('S091', 'Mastering ScaNN for Cost Efficiency', 'Join this session to explore how ScaNN can be used for Cost Efficiency. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S091', 'DevOps Engineer', '2026-04-09', '14:00:00', '14:00:00', 'Auditorium', 100, 21, 'AI & Machine Learning', 'Technology'),
('S092', 'Mastering BigQuery for Scale', 'Join this session to explore how BigQuery can be used for Scale. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Advanced', 'http://sessions.com/S092', 'Cloud Architect', '2026-04-09', '16:00:00', '16:00:00', 'Room C', 200, 67, 'Databases', 'Manufacturing'),
('S093', 'Exploring AlloyDB for Performance', 'Join this session to explore how AlloyDB can be used for Performance. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S093', 'Security Engineer', '2026-04-09', '10:00:00', '10:00:00', 'Room B', 200, 171, 'Databases', 'Finance'),
('S094', 'Deploying Cloud SQL for Modernization', 'Join this session to explore how Cloud SQL can be used for Modernization. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Advanced', 'http://sessions.com/S094', 'Product Manager', '2026-04-09', '14:00:00', '14:00:00', 'Room B', 100, 71, 'DevOps', 'Healthcare'),
('S095', 'Building Spanner for Analytics', 'Join this session to explore how Spanner can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S095', 'Product Manager', '2026-04-10', '15:00:00', '15:00:00', 'Room B', 500, 334, 'Databases', 'Manufacturing'),
('S096', 'Deep Dive into Vertex AI for Analytics', 'Join this session to explore how Vertex AI can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Intermediate', 'http://sessions.com/S096', 'DevOps Engineer', '2026-04-11', '14:00:00', '14:00:00', 'Room C', 100, 27, 'AI & Machine Learning', 'Healthcare'),
('S097', 'Deploying PostgreSQL for Analytics', 'Join this session to explore how PostgreSQL can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Breakouts', 'Beginner', 'http://sessions.com/S097', 'Cloud Architect', '2026-04-09', '11:00:00', '11:00:00', 'Room C', 50, 36, 'Databases', 'Healthcare'),
('S098', 'Architecting Kubernetes for Reliability', 'Join this session to explore how Kubernetes can be used for Reliability. We will cover best practices, real-world use cases, and advanced configuration options.', 'Lightning Talks', 'Beginner', 'http://sessions.com/S098', 'Cloud Architect', '2026-04-10', '11:00:00', '11:00:00', 'Room A', 50, 29, 'AI & Machine Learning', 'Finance'),
('S099', 'Scaling Cloud SQL for Analytics', 'Join this session to explore how Cloud SQL can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Beginner', 'http://sessions.com/S099', 'DevOps Engineer', '2026-04-09', '09:00:00', '09:00:00', 'Auditorium', 50, 21, 'Security', 'Finance'),
('S100', 'Deploying PostgreSQL for Analytics', 'Join this session to explore how PostgreSQL can be used for Analytics. We will cover best practices, real-world use cases, and advanced configuration options.', 'Keynotes', 'Advanced', 'http://sessions.com/S100', 'DevOps Engineer', '2026-04-11', '09:00:00', '09:00:00', 'Room B', 500, 141, 'Databases', 'Technology') ON CONFLICT (session_id) DO NOTHING;

3.5 Attribuer des intervenants aux sessions

-- Insert Session-Speaker Mappings
INSERT INTO public.session_speaker_mapping (session_id, speaker_id) VALUES
('S001', 6),
('S001', 19),
('S002', 6),
('S002', 11),
('S003', 14),
('S003', 10),
('S004', 5),
('S004', 8),
('S005', 12),
('S005', 9),
('S006', 14),
('S007', 14),
('S007', 3),
('S008', 2),
('S008', 4),
('S009', 9),
('S009', 13),
('S010', 2),
('S010', 3),
('S011', 12),
('S012', 13),
('S013', 11),
('S014', 8),
('S015', 16),
('S016', 4),
('S016', 18),
('S017', 12),
('S018', 10),
('S019', 15),
('S020', 20),
('S020', 7),
('S021', 2),
('S021', 11),
('S022', 15),
('S022', 7),
('S023', 4),
('S023', 19),
('S024', 11),
('S025', 11),
('S025', 10),
('S026', 11),
('S027', 20),
('S028', 20),
('S028', 3),
('S029', 11),
('S029', 20),
('S030', 19),
('S030', 6),
('S031', 20),
('S031', 14),
('S032', 3),
('S033', 4),
('S033', 8),
('S034', 14),
('S035', 12),
('S035', 4),
('S036', 18),
('S037', 6),
('S038', 9),
('S038', 7),
('S039', 7),
('S040', 17),
('S041', 17),
('S042', 16),
('S042', 2),
('S043', 10),
('S043', 13),
('S044', 7),
('S045', 10),
('S046', 1),
('S046', 20),
('S047', 17),
('S048', 9),
('S048', 6),
('S049', 4),
('S049', 3),
('S050', 8),
('S050', 16),
('S051', 18),
('S051', 10),
('S052', 4),
('S053', 5),
('S053', 13),
('S054', 17),
('S055', 1),
('S056', 1),
('S056', 15),
('S057', 7),
('S058', 15),
('S059', 15),
('S059', 8),
('S060', 20),
('S060', 10),
('S061', 16),
('S062', 18),
('S062', 11),
('S063', 10),
('S064', 6),
('S065', 10),
('S066', 8),
('S066', 9),
('S067', 16),
('S067', 14),
('S068', 13),
('S068', 11),
('S069', 16),
('S070', 8),
('S071', 3),
('S072', 10),
('S072', 1),
('S073', 10),
('S073', 8),
('S074', 20),
('S074', 14),
('S075', 5),
('S076', 1),
('S076', 11),
('S077', 3),
('S078', 20),
('S078', 12),
('S079', 17),
('S079', 9),
('S080', 10),
('S081', 2),
('S082', 15),
('S082', 11),
('S083', 7),
('S083', 1),
('S084', 8),
('S085', 8),
('S085', 18),
('S086', 5),
('S087', 5),
('S088', 19),
('S088', 18),
('S089', 18),
('S090', 8),
('S091', 4),
('S092', 14),
('S093', 20),
('S093', 17),
('S094', 16),
('S094', 9),
('S095', 11),
('S095', 4),
('S096', 14),
('S096', 11),
('S097', 13),
('S097', 19),
('S098', 3),
('S099', 11),
('S099', 18),
('S100', 18),
('S100', 13) ON CONFLICT DO NOTHING;

3.6 Associer des thèmes à des sessions

-- Insert Session-Topic Mappings
INSERT INTO public.session_topic_mapping (session_id, topic) VALUES
('S001', 'Security'),
('S002', 'AI & Machine Learning'),
('S003', 'Cloud Architecture'),
('S003', 'AI & Machine Learning'),
('S004', 'Databases'),
('S004', 'DevOps'),
('S005', 'DevOps'),
('S005', 'Cloud Architecture'),
('S006', 'Security'),
('S006', 'AI & Machine Learning'),
('S007', 'Cloud Architecture'),
('S007', 'Databases'),
('S008', 'AI & Machine Learning'),
('S008', 'Databases'),
('S009', 'Security'),
('S009', 'DevOps'),
('S010', 'AI & Machine Learning'),
('S010', 'Cloud Architecture'),
('S011', 'Security'),
('S012', 'Databases'),
('S013', 'Cloud Architecture'),
('S014', 'Databases'),
('S015', 'DevOps'),
('S015', 'Security'),
('S016', 'AI & Machine Learning'),
('S016', 'Databases'),
('S017', 'Cloud Architecture'),
('S018', 'AI & Machine Learning'),
('S019', 'AI & Machine Learning'),
('S020', 'Security'),
('S021', 'Security'),
('S021', 'Databases'),
('S022', 'Security'),
('S022', 'Databases'),
('S023', 'DevOps'),
('S023', 'AI & Machine Learning'),
('S024', 'AI & Machine Learning'),
('S025', 'Databases'),
('S026', 'DevOps'),
('S027', 'DevOps'),
('S027', 'Cloud Architecture'),
('S028', 'DevOps'),
('S029', 'Databases'),
('S029', 'Security'),
('S030', 'DevOps'),
('S030', 'Security'),
('S031', 'Databases'),
('S032', 'Databases'),
('S033', 'Databases'),
('S034', 'AI & Machine Learning'),
('S035', 'Security'),
('S035', 'AI & Machine Learning'),
('S036', 'AI & Machine Learning'),
('S036', 'Databases'),
('S037', 'DevOps'),
('S037', 'Databases'),
('S038', 'Cloud Architecture'),
('S038', 'Security'),
('S039', 'Security'),
('S040', 'Security'),
('S040', 'AI & Machine Learning'),
('S041', 'AI & Machine Learning'),
('S042', 'Databases'),
('S043', 'AI & Machine Learning'),
('S043', 'Databases'),
('S044', 'Databases'),
('S044', 'DevOps'),
('S045', 'DevOps'),
('S045', 'Cloud Architecture'),
('S046', 'Cloud Architecture'),
('S046', 'Security'),
('S047', 'Cloud Architecture'),
('S047', 'Security'),
('S048', 'Databases'),
('S049', 'AI & Machine Learning'),
('S049', 'Databases'),
('S050', 'Cloud Architecture'),
('S050', 'AI & Machine Learning'),
('S051', 'DevOps'),
('S051', 'Databases'),
('S052', 'Databases'),
('S053', 'DevOps'),
('S053', 'Databases'),
('S054', 'Security'),
('S054', 'DevOps'),
('S055', 'AI & Machine Learning'),
('S056', 'Security'),
('S056', 'DevOps'),
('S057', 'AI & Machine Learning'),
('S058', 'Security'),
('S058', 'AI & Machine Learning'),
('S059', 'Security'),
('S060', 'AI & Machine Learning'),
('S060', 'DevOps'),
('S061', 'Security'),
('S061', 'AI & Machine Learning'),
('S062', 'DevOps'),
('S063', 'Security'),
('S063', 'DevOps'),
('S064', 'Databases'),
('S064', 'AI & Machine Learning'),
('S065', 'DevOps'),
('S066', 'AI & Machine Learning'),
('S066', 'Cloud Architecture'),
('S067', 'AI & Machine Learning'),
('S068', 'Security'),
('S069', 'Cloud Architecture'),
('S069', 'DevOps'),
('S070', 'AI & Machine Learning'),
('S071', 'AI & Machine Learning'),
('S071', 'DevOps'),
('S072', 'DevOps'),
('S072', 'Security'),
('S073', 'Security'),
('S074', 'AI & Machine Learning'),
('S074', 'Databases'),
('S075', 'AI & Machine Learning'),
('S076', 'DevOps'),
('S077', 'DevOps'),
('S077', 'AI & Machine Learning'),
('S078', 'Databases'),
('S079', 'Databases'),
('S080', 'AI & Machine Learning'),
('S080', 'Databases'),
('S081', 'Databases'),
('S082', 'Databases'),
('S083', 'Security'),
('S084', 'Security'),
('S084', 'AI & Machine Learning'),
('S085', 'AI & Machine Learning'),
('S085', 'Databases'),
('S086', 'DevOps'),
('S087', 'Databases'),
('S088', 'Security'),
('S088', 'Databases'),
('S089', 'Databases'),
('S090', 'DevOps'),
('S091', 'Databases'),
('S092', 'Databases'),
('S093', 'Security'),
('S093', 'AI & Machine Learning'),
('S094', 'Cloud Architecture'),
('S095', 'Cloud Architecture'),
('S095', 'AI & Machine Learning'),
('S096', 'AI & Machine Learning'),
('S097', 'Security'),
('S097', 'AI & Machine Learning'),
('S098', 'Databases'),
('S098', 'Security'),
('S099', 'Cloud Architecture'),
('S100', 'DevOps'),
('S100', 'Cloud Architecture') ON CONFLICT DO NOTHING;;

3.7 Attribuer des participants à des sessions

-- Insert Attendee-Session Mappings
INSERT INTO public.attendees_sessions (username, session_id) VALUES
('user_1', 'S100'),
('user_1', 'S008'),
('user_1', 'S060'),
('user_1', 'S090'),
('user_1', 'S057'),
('user_2', 'S086'),
('user_2', 'S033'),
('user_2', 'S006'),
('user_2', 'S043'),
('user_2', 'S050'),
('user_3', 'S066'),
('user_3', 'S099'),
('user_4', 'S004'),
('user_4', 'S043'),
('user_4', 'S092'),
('user_4', 'S033'),
('user_4', 'S074'),
('user_5', 'S014'),
('user_5', 'S088'),
('user_5', 'S093'),
('user_6', 'S075'),
('user_6', 'S033'),
('user_7', 'S014'),
('user_7', 'S021'),
('user_7', 'S047'),
('user_8', 'S051'),
('user_8', 'S081'),
('user_9', 'S048'),
('user_9', 'S100'),
('user_10', 'S037'),
('user_10', 'S059'),
('user_10', 'S083'),
('user_11', 'S007'),
('user_11', 'S099'),
('user_11', 'S054'),
('user_12', 'S006'),
('user_12', 'S046'),
('user_12', 'S077'),
('user_12', 'S032'),
('user_13', 'S020'),
('user_13', 'S029'),
('user_13', 'S054'),
('user_14', 'S052'),
('user_14', 'S070'),
('user_14', 'S028'),
('user_15', 'S054'),
('user_15', 'S050'),
('user_15', 'S025'),
('user_15', 'S066'),
('user_15', 'S081'),
('user_16', 'S099'),
('user_16', 'S073'),
('user_16', 'S027'),
('user_16', 'S058'),
('user_17', 'S092'),
('user_17', 'S089'),
('user_17', 'S076'),
('user_17', 'S062'),
('user_18', 'S083'),
('user_18', 'S094'),
('user_18', 'S097'),
('user_18', 'S031'),
('user_18', 'S040'),
('user_19', 'S078'),
('user_19', 'S072'),
('user_19', 'S049'),
('user_19', 'S017'),
('user_19', 'S084'),
('user_20', 'S023'),
('user_20', 'S003'),
('user_20', 'S016'),
('user_20', 'S068'),
('user_21', 'S071'),
('user_21', 'S058'),
('user_21', 'S043'),
('user_21', 'S079'),
('user_21', 'S067'),
('user_22', 'S076'),
('user_22', 'S038'),
('user_22', 'S049'),
('user_22', 'S033'),
('user_22', 'S070'),
('user_23', 'S032'),
('user_23', 'S099'),
('user_24', 'S022'),
('user_24', 'S065'),
('user_24', 'S060'),
('user_24', 'S084'),
('user_25', 'S077'),
('user_25', 'S080'),
('user_25', 'S097'),
('user_25', 'S010'),
('user_26', 'S063'),
('user_26', 'S052'),
('user_26', 'S086'),
('user_27', 'S054'),
('user_27', 'S094'),
('user_27', 'S018'),
('user_27', 'S061'),
('user_27', 'S052'),
('user_28', 'S064'),
('user_28', 'S070'),
('user_28', 'S050'),
('user_29', 'S009'),
('user_29', 'S012'),
('user_30', 'S058'),
('user_30', 'S056'),
('user_30', 'S072'),
('user_30', 'S093'),
('user_30', 'S045'),
('user_31', 'S001'),
('user_31', 'S094'),
('user_31', 'S065'),
('user_31', 'S031'),
('user_32', 'S048'),
('user_32', 'S011'),
('user_32', 'S065'),
('user_33', 'S021'),
('user_33', 'S081'),
('user_33', 'S063'),
('user_34', 'S068'),
('user_34', 'S026'),
('user_35', 'S044'),
('user_35', 'S054'),
('user_36', 'S023'),
('user_36', 'S051'),
('user_36', 'S100'),
('user_37', 'S047'),
('user_37', 'S053'),
('user_37', 'S057'),
('user_37', 'S048'),
('user_37', 'S080'),
('user_38', 'S003'),
('user_38', 'S038'),
('user_38', 'S046'),
('user_38', 'S005'),
('user_38', 'S076'),
('user_39', 'S046'),
('user_39', 'S020'),
('user_39', 'S043'),
('user_39', 'S002'),
('user_39', 'S100'),
('user_40', 'S019'),
('user_40', 'S098'),
('user_40', 'S053'),
('user_40', 'S007'),
('user_41', 'S100'),
('user_41', 'S032'),
('user_41', 'S048'),
('user_41', 'S064'),
('user_42', 'S086'),
('user_42', 'S030'),
('user_42', 'S049'),
('user_43', 'S033'),
('user_43', 'S008'),
('user_43', 'S049'),
('user_43', 'S093'),
('user_44', 'S078'),
('user_44', 'S071'),
('user_44', 'S067'),
('user_44', 'S012'),
('user_45', 'S043'),
('user_45', 'S056'),
('user_45', 'S079'),
('user_45', 'S062'),
('user_45', 'S039'),
('user_46', 'S054'),
('user_46', 'S031'),
('user_46', 'S018'),
('user_47', 'S035'),
('user_47', 'S100'),
('user_47', 'S083'),
('user_47', 'S080'),
('user_47', 'S037'),
('user_48', 'S097'),
('user_48', 'S026'),
('user_48', 'S065'),
('user_48', 'S030'),
('user_48', 'S074'),
('user_49', 'S022'),
('user_49', 'S089'),
('user_49', 'S038'),
('user_49', 'S047'),
('user_49', 'S073'),
('user_50', 'S074'),
('user_50', 'S033'),
('user_50', 'S069'),
('user_50', 'S036') ON CONFLICT DO NOTHING;

Vous devriez obtenir un résultat indiquant que l'exécution a réussi. Vérifiez que les tables sont remplies. Par exemple, vérifiez que la table des sessions contient 100 sessions :

SELECT count(*) from public.session;

7. Activer le moteur de requêtes d'IA

Avant d'utiliser les fonctions d'IA, vous devez activer le moteur de requête d'IA dans votre base de données.

Exécutez le code SQL suivant dans AlloyDB Studio (contexte conference_db) :

CREATE EXTENSION IF NOT EXISTS google_ml_integration CASCADE;
ALTER DATABASE conference_db SET google_ml_integration.enable_ai_query_engine = 'on';

Vérifiez que l'extension est activée :

SELECT extversion FROM pg_extension WHERE extname = 'google_ml_integration';

Le résultat attendu doit afficher 1.5.9 ou une valeur supérieure.

8. Utiliser les fonctions d'IA (opérateurs)

Utilisons maintenant le filtrage et le scoring sémantiques basés sur l'IA avec les fonctions ai.if et ai.rank.

Filtrage sémantique avec ai.if

La correspondance textuelle standard peut manquer des sessions si le mot clé exact n'est pas présent. Trouvons des sessions sur l'IA générative à l'aide de ai.if.

Exécutez le code SQL suivant dans AlloyDB Studio :

SELECT session_name, full_description
FROM public.session
WHERE
  ai.if(
    prompt => 'Is this session description about Generative AI, LLMs or AI agents? ' || full_description
  );

Vous devriez voir les sessions pertinentes dans les résultats, même si elles utilisent une terminologie différente.

Évaluation sémantique avec ai.rank

Évaluons les sessions en fonction de leur accessibilité pour les débutants. Nous pouvons utiliser ai.rank pour le score sémantique.

Exécutez le code SQL suivant :

SELECT session_name, full_description, 
  ai.rank(
    prompt => 'On a scale of 0 to 1, how suitable is this session for a beginner? 1 being very suitable, 0 being advanced. Session: ' || full_description
  ) as beginner_friendly_score
FROM public.session
ORDER BY beginner_friendly_score DESC;

Cela permet de créer des recommandations personnalisées pour les participants à la conférence en fonction de leur profil.

9. Implémenter la recherche hybride

La recherche hybride combine la précision de la recherche par mot clé (lexicale) avec la compréhension contextuelle de la recherche vectorielle (sémantique). Nous allons créer des index pour les deux et utiliser la fusion de classement réciproque (RRF) pour fusionner les résultats.

Activer l'analyse

Assurez-vous que les extensions scann et rum sont activées :

CREATE EXTENSION IF NOT EXISTS alloydb_scann;
CREATE EXTENSION IF NOT EXISTS rum;

Créer des index

Étant donné que notre schéma utilise une colonne générée pour les embeddings (full_description_embedding), ils sont automatiquement calculés lors de l'insertion. Il nous suffit de créer les index appropriés pour une récupération rapide.

Créez un index RUM (index FTS) pour la recherche de mots clés en texte intégral et un index ScaNN pour la recherche rapide de similarité vectorielle.

Exécutez le code SQL suivant dans AlloyDB Studio :

-- Create RUM index for text search on session descriptions
CREATE INDEX session_text_idx ON public.session USING RUM (description_tsvector rum_tsvector_ops);

-- Create ScaNN index for vector search on session embeddings
CREATE INDEX session_vector_idx ON public.session
  USING scann (full_description_embedding cosine)
  WITH (num_leaves=10);

Exécuter une recherche hybride avec RRF

Nous allons maintenant rechercher une session sur fast similarity search, en nous concentrant sur ScaNN d'AlloyDB.

Nous utiliserons la recherche sémantique et la recherche par mots clés, puis nous fusionnerons les résultats à l'aide de la RRF.

Exécutez le code SQL suivant :

-- Enable preview features for the AI Query Engine if not already set
SET google_ml_integration.enable_preview_ai_functions = true;
SELECT s.session_id, s.session_name, s.full_description
FROM public.session s
JOIN ai.hybrid_search(
  search_inputs => ARRAY[
      '{
        "data_type": "vector",
        "table_name": "session",
        "key_column": "session_id",
        "vec_column": "full_description_embedding",
        "distance_operator": "public.<=>",
        "limit": 5,
        "query_vector": "ai.embedding(''text-embedding-005'', ''fast similarity search'')::vector"
      }'::JSONB,
      '{
        "data_type": "text",
        "table_name": "session",
        "key_column": "session_id",
        "text_column": "description_tsvector",
        "limit": 5,
        "ranking_function": "<=>",
        "query_text_input": "ScaNN"
      }'::JSONB
  ]
) AS search_results
on s.session_id = search_results.id;

Vous devriez voir une recherche de similarité de sessions pertinentes, la plupart des résultats affichant des sessions associées à ScaNN.

10. Utiliser QueryData dans AlloyDB Studio

AlloyDB AI vous permet d'utiliser QueryData pour générer des instructions SQL précises et prévisibles à partir d'entrées en langage naturel. Dans cette section, vous allez apprendre à créer un contexte QueryData (modèles) et à le tester directement dans AlloyDB Studio.

Créer un contexte QueryData

Le contexte QueryData est un fichier JSON contenant des modèles de requête et des facettes qui fournissent les données et les instructions nécessaires au modèle d'IA pour utiliser les requêtes SQL ou les parties de requête SQL appropriées.

Examinons un exemple de contexte JSON conçu pour notre schéma de conférence. Notez que nous nous assurons que les noms de tables correspondent à notre schéma (par exemple, public.session).

Voici le contenu JSON que nous allons utiliser. Stockez-le dans un fichier .json local sur votre ordinateur portable ou de bureau :

{
  "templates": [
    {
      "nlQuery": "Advanced sessions that are almost full",
      "sql": "SELECT session_name, remaining_capacity, session_date, session_start_time FROM session WHERE learning_level = 'Advanced' AND remaining_capacity > 0 AND remaining_capacity < 5 ORDER BY remaining_capacity ASC",
      "intent": "Advanced sessions that are almost full (remaining capacity less than 5)",
      "manifest": "Advanced sessions that are almost full (remaining capacity less than 5)",
      "parameterized": {
        "parameterized_intent": "$1 sessions that are almost full (remaining capacity less than 5)",
        "parameterized_sql": "SELECT session_name, remaining_capacity, session_date, session_start_time FROM public.session WHERE learning_level = $1 AND remaining_capacity > 0 AND remaining_capacity < 5 ORDER BY remaining_capacity ASC"
      }
    },
    {
      "nlQuery": "Find sessions about Gemini",
      "sql": "SELECT name, format, session_date, description FROM ((SELECT s1.session_name AS name, s1.session_format AS format, s1.session_date AS session_date, s1.full_description AS description, (s1.full_description_embedding <=> public.embedding('text-embedding-005', 'Gemini')::public.vector) AS distance FROM public.session s1 ORDER BY distance LIMIT 10) UNION ALL (SELECT s2.session_name AS name, s2.session_format AS format, s2.session_date AS session_date, t.topic_desc AS description, (t.embedding <=> public.embedding('text-embedding-005', 'Gemini')::public.vector) AS distance FROM public.session s2 INNER JOIN public.session_topic_mapping stm ON s2.session_id = stm.session_id INNER JOIN public.session_topic t ON stm.topic = t.topic ORDER BY distance LIMIT 10)) AS combined_results ORDER BY distance LIMIT 10",
      "intent": "Find sessions about Gemini",
      "manifest": "Find sessions about a given topic",
      "parameterized": {
        "parameterized_intent": "Find sessions about $1",
        "parameterized_sql": "SELECT name, format, session_date, description FROM ((SELECT s1.session_name AS name, s1.session_format AS format, s1.session_date AS session_date, s1.full_description AS description, (s1.full_description_embedding <=> public.embedding('text-embedding-005', '$1')::public.vector) AS distance FROM public.session s1 ORDER BY distance LIMIT 10) UNION ALL (SELECT s2.session_name AS name, s2.session_format AS format, s2.session_date AS session_date, t.topic_desc AS description, (t.embedding <=> public.embedding('text-embedding-005', '$1')::public.vector) AS distance FROM public.session s2 INNER JOIN public.session_topic_mapping stm ON s2.session_id = stm.session_id INNER JOIN public.session_topic t ON stm.topic = t.topic ORDER BY distance LIMIT 10)) AS combined_results ORDER BY distance LIMIT 10"
      }
    }
  ]
}

Remarque : Dans le deuxième modèle, nous utilisons full_description_embedding, qui est le nom de la colonne dans notre schéma.

Charger le contexte QueryData dans AlloyDB Studio

Pour utiliser notre contexte "Interroger les données", nous devons l'importer dans notre base de données via AlloyDB Studio.

1. Ouvrez AlloyDB Studio dans la console Google Cloud.
2. En bas du panneau de gauche, vous verrez Ensembles de contexte et trois points.
3. Cliquez dessus, puis sélectionnez Créer un ensemble de contexte.
4. Remplissez la boîte de dialogue :
   • Nom : conference_context
   • Description : Conference Sessions QueryData Context
   • Importer le fichier de contexte : importez le fichier JSON créé avec le contenu ci-dessus.
5. Enregistrer le conten

La création du contexte peut prendre jusqu'à cinq minutes.

Tester le contexte QueryData

Une fois le fichier importé, vous pouvez le tester directement dans AlloyDB Studio.

1. Cliquez sur les trois points à côté du contexte que vous venez de créer, puis sélectionnez Tester l'ensemble de contextes (ou utilisez le bouton Gemini dans l'éditeur de requête et sélectionnez ce contexte).
2. Dans le prompt de génération SQL de Gemini, saisissez une requête en langage naturel qui correspond à notre modèle :

   Advanced sessions that are almost full
   

3. Générez le code SQL et vérifiez qu'il correspond au modèle que nous avons fourni.
4. Essayez une requête paramétrée en remplaçant "Avancé" par "Débutant" dans votre requête en langage naturel et voyez si elle s'adapte.

11. Effectuer un nettoyage

Supprimer un cluster AlloyDB

Pour éviter que les ressources créées lors de cet atelier de programmation ne soient facturées en permanence sur votre compte Google Cloud, supprimez-les.

Exécutez les commandes suivantes dans Cloud Shell pour supprimer le cluster AlloyDB (l'instance sera également supprimée) :

export REGION=us-central1
export ADBCLUSTER=alloydb-next26-ai-demo-01

echo "=> Deleting AlloyDB Cluster (${ADBCLUSTER})..."
gcloud alloydb clusters delete $ADBCLUSTER --region=$REGION --force

Remarque : L'exécution de la commande prend trois à cinq minutes.

Si vous avez créé un projet spécifiquement pour cet atelier de programmation, vous pouvez également le supprimer entièrement.

Supprimer la plage d'adresses IP et l'appairage de VPC

⚠️ Avertissement pour les projets partagés : Le script de nettoyage ci-dessous supprime également la connexion d'appairage VPC Service Networking (servicenetworking.googleapis.com). Cette opération est parfaitement sûre et recommandée si vous utilisez un projet Codelab isolé et jetable. Toutefois, si vous exécutez cette opération dans un projet Google Cloud partagé ou de production, la suppression de cet appairage déconnectera instantanément les autres bases de données gérées (comme Cloud SQL ou Memorystore) qui partagent le même VPC. Si vous avez d'autres services qui dépendent de ce VPC, ignorez cette étape.

PROJECT_ID=$(gcloud config get-value project)
echo "=> Deleting Service Networking VPC Peering..."
gcloud compute networks peerings delete servicenetworking-googleapis-com \
    --network=default \
    --project=${PROJECT_ID} --quiet || true

echo "=> Deleting Allocated IP Range for Managed Services..."
gcloud compute addresses delete psa-range \
    --global \
    --project=${PROJECT_ID} --quiet || true

Supprimer un fichier local

Supprimez le fichier JSON local créé à l'étape Work with QueryData in AlloyDB Studio.

12. Félicitations

Félicitations ! Vous avez exploré plusieurs fonctionnalités d'AlloyDB AI à l'aide d'un scénario unifié de sessions de conférence.

Points abordés

• Comment activer et utiliser les fonctions d'IA (ai.if, ai.rank) pour le filtrage et la notation sémantiques.
• Découvrez comment implémenter la recherche hybride à l'aide des index ScaNN (vectoriel) et RUM (texte) avec la fusion de classement réciproque (RRF).
• Découvrez comment utiliser QueryData dans AlloyDB Studio pour générer du code SQL prévisible à partir du langage naturel.

Étapes suivantes

• Consultez la documentation AlloyDB AI.
• En savoir plus sur pgvector et ScaNN

Documents de référence

• Présentation d'AlloyDB AI
• Requêtes en langage naturel