Answer App

A production-ready Retrieval-Augmented Generation (RAG) server that uses Vertex AI Search and the Discovery Engine API to serve the Answer method - a conversational search experience with generative answers grounded on document data.

• 🤖 Fully-managed RAG pipeline: Stateful multi-turn conversational search with generative answers
• ⚡ Production-capable performance: Autoscaling, concurrency, and regional redundancy via multiple Cloud Run services
• 🔗 Integration flexibility: Authenticated external HTTPS endpoint using Google-managed TLS
• 📊 Resource observability: Single-pane-of-glass Cloud Monitoring dashboard with customizable metrics and alerts
• 🔍 Explainable and debuggable results: Investigate generative answers using the full RAG pipeline results logged to BigQuery
• 📈 Data-driven LLM-ops: Tune the conversational search agent using question/answer pairs labelled with user feedback
• 🛡️ Identity-Aware Proxy: Secure access control
• 👤 Google OAuth: Personalized sessions with user authentication
• 🚀 Automated deployments: One-click install and uninstall with Terraform and Cloud Build

Architecture

• The Global External Application Load Balancer provides planet-scale availability
• The load balancer backend service interfaces with regional serverless network endpoint group backends composed of Cloud Run services
  • Zonal failover: Cloud Run services replicate across multiple zones within a Compute region to prevent outages for a single zonal failure
  • Autoscaling: add/remove instances to match demand and maintain a minimum instance count for high availability
  • Concurrency: instances process multiple requests simultaneously
  • Regional redundancy: services can span multiple regions to optimize latency and optionally deliver higher availability in case of regional outages.
• The Vertex AI Search Search App and Data Store automate document preparation for semantic search and retrieval
• The Conversational Search Service (the interface for the Answer method) uses Gemini-based answer generation models to power grounded generative answers
• The application asynchronously writes the full session data and user feedback responses to BigQuery for offline analysis

Quick Start

Prerequisites

• Google Cloud Project with Owner permissions
• Terraform and gcloud CLI installed

See detailed deployment prerequisites →

Installation

1. Configure OAuth for user authentication
   📖 Complete OAuth setup guide →

2. Deploy the application

   source scripts/install.sh

3. Enable Vertex AI Agent Builder in the Cloud Console and import your documents

4. Configure Identity-Aware Proxy to secure access

📖 View complete installation guide →

Documentation

Installation & Deployment

• ✅ Prerequisites - Environment setup
• 🔐 OAuth Setup Guide - Step-by-step OAuth client configuration
• 📋 Deployment - Deployment and Post-deployment steps

Development

• 🧪 Development Guide - Local development, testing, and Docker usage
• 📖 API Reference - Answer method configuration options
• 🏷️ Version Management - Automated semantic release and versioning

Infrastructure

• 🏗️ Terraform Overview - General Terraform patterns and best practices (reusable)
• 🚀 Bootstrap Process - Initial project setup and service accounts
• ☁️ Cloud Build Automation - Automated deployments and CI/CD
• 🔄 Rollbacks - Rolling back deployments and managing revisions
• ⚙️ Infrastructure Changes - Applying infrastructure-only changes
• 🛠️ Helper Scripts - Automation scripts reference

Troubleshooting

• ❗ Known Issues - Common problems and solutions

Uninstall

Remove all resources:

source scripts/uninstall.sh

Development

This project uses:

• Python 3.13+ with Poetry for dependency management
• FastAPI backend with an example Streamlit frontend
• Terraform for infrastructure as code
• pytest for testing

📖 Full development guide →

Repository Structure

answer-app/
├── src/
│   ├── answer_app/          # FastAPI backend service
│   ├── client/              # Streamlit frontend application
│   └── package_scripts/     # Helper scripts (OAuth secrets)
├── terraform/
│   ├── bootstrap/           # Initial project setup
│   ├── main/                # Main infrastructure deployment
│   └── modules/             # Reusable Terraform modules
├── docs/                    # Modular documentation
│   ├── installation/        # Setup guides
│   ├── infrastructure/      # Infrastructure documentation
│   ├── development/         # Development & API docs
│   └── troubleshooting/     # Known issues & solutions
├── scripts/                 # Automation scripts
├── tests/                   # Unit tests
└── assets/                  # Documentation screenshots

License

This project is licensed under the MIT License - see the LICENSE file for details.