GCP RAG Chatbot

This was originally pieced together using publically available information. Maybe someday I will relase the guide I wrote on how to set up the infrastructure once it is common knowledge.

A modular Retrieval-Augmented Generation (RAG) chatbot built on Google Cloud Platform (GCP). This project provides tools for document ingestion, embedding, and chat interaction using a custom agent backed by Google Cloud services.

🧱 Project Structure

gcp-rag-chatbot-main/
├── gcs_chat_app/       # Streamlit chat interface
├── gcs_embed/          # Document embedding pipeline
├── rag_agent/          # RAG agent logic and utilities
├── .gitignore
├── LICENSE
└── README.md

🚀 Components

gcs_chat_app

• Serves chat requests (likely with FastAPI)
• Imports documents from GCS
• Dockerized with Cloud Build support

gcs_embed

• Converts documents into vector embeddings
• Cloud Build and Workflows-ready
• Integrates with GCP storage and possibly Vertex AI

rag_agent

• Core RAG agent logic
• Tools for managing conversation state and node logic

🛠️ Setup Instructions

Prerequisites

• Python 3.9+
• Docker
• Google Cloud SDK (gcloud)
• Enabled GCP services: Cloud Storage, Artifact Registry, Vertex AI (if used)

Clone the Repository

git clone https://github.com/your-org/gcp-rag-chatbot.git
cd gcp-rag-chatbot

Environment Variables

Copy .env.example in gcs_chat_app/ and gcs_embed/ to .env and configure as needed:

cp gcs_chat_app/.env.example gcs_chat_app/.env
cp gcs_embed/.env.example gcs_embed/.env

Build and Deploy with Cloud Build (Optional)

gcloud builds submit --config=gcs_chat_app/cloudbuild.yaml

Or for embeddings:

gcloud builds submit --config=gcs_embed/cloudbuild.yaml

🧠 How it Works

1. Documents are uploaded or fetched from GCS.
2. gcs_embed embeds these documents and stores vector data.
3. gcs_chat_app accepts chat input, calls the RAG agent, and returns responses.
4. rag_agent uses tools and embeddings to retrieve and answer questions.

🧪 Local Development

To run the chat app locally:

cd gcs_chat_app
uvicorn app:app --reload

To test embeddings:

cd gcs_embed
python embed.py

TODO

• Find new open source data source for training materials and them to the repo
• Modify embeddings functionality to fit new data source
• Share my guide on how to set up the infrastructure (Once it is common knowledge)

📄 License

MIT License. See the LICENSE file for details.

---

This project is designed to be production-ready on GCP and easily extensible with new document sources, models, or UI interfaces. Contributions welcome!