---

🔍 Gemini RAG Assistant

A production-ready RAG (Retrieval-Augmented Generation) web application that leverages Google Gemini Pro for question answering using user-uploaded PDFs or CSVs. Built with FastAPI, LangChain, Pinecone, and LangSmith for robust LLMOps, and deployed using Docker, GitHub Actions, and Render.

---

🚀 Features

• 🔮 Uses Gemini 2.0 Flash for conversational LLM responses
• 📚 Upload PDF or CSV files to dynamically extend knowledge
• 🧠 Context storage using Pinecone Vector Database
• 🧩 Built with LangChain for easy chaining, Memory and RAG architecture
• 🧪 Tracks sessions, queries, cost, latency via LangSmith
• 🛠️ Built with FastAPI + Jinja2 templates for the web UI
• 🔐 Prevents duplicate uploads using SHA256 content hashing
• 📦 Automatically tested and deployed via GitHub Actions
• 🐳 Packaged in a Docker container and deployed to Render

---

🖼️ UI Screenshots

• Home Page

---

## 🧱 Project Structure

.
├── app/
│   ├── main.py                                        # FastAPI backend logic
│   ├── templates/                                     # Jinja2 HTML templates
│   ├── static/                                        # Static CSS/JS/images
│   ├ ____________________├── filenames in mongodb     # Tracks uploaded filenames in mongodb
│   ├                     ├── hashes in mongodb        # Tracks file hashes to prevent duplicates in mongodb
│   └── temp_uploads/                                  # Temporary storage during processing
├── requirements.txt
├── .env                                               # API keys and secrets (not committed)
├── Dockerfile
├── .github/
│   └── workflows/
│       └── deploy.yaml                                # GitHub Actions CI/CD pipeline
├-------------------------├ Pinecone Vector DB         # Storing embeddings in Vector DB
└── README.md

---

🧠 LLM Stack Overview

Component	Tech Used
Language Model	Gemini 2.0 Flash (ChatGoogleGenerativeAI)
Embeddings	GoogleGenerativeAIEmbeddings
Vector Store	Pinecone
Chunking	LangChain's RecursiveCharacterTextSplitter
RAG Chain	ConversationalRetrievalChain
Memory	ConversationBufferMemory
Tracing/Monitoring	LangSmith

---

📊 LangSmith Integration

LangSmith is used to trace and monitor:

• ✅ Query and response pairs
• 🕒 Latency per interaction
• 💰 Token usage and cost
• 🧵 Session chat history

Make sure your .env file contains:

LANGSMITH_API_KEY=your_langsmith_key
LANGSMITH_TRACING=true

---

🔧 Requirements

Install all dependencies via:

pip install -r requirements.txt

---

📄 .env Configuration

Make sure you create a .env file in the root directory with:

GOOGLE_API_KEY=your_google_api_key
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_API_ENV=us-east-1
LANGSMITH_API_KEY=your_langsmith_api_key
LANGSMITH_TRACING=true

---

🔬 Testing & CI/CD

This project uses GitHub Actions to:

1. Run 3 test cases for app health check and upload/chat routes
2. On successful tests, build a Docker image
3. Push the image to DockerHub automatically

Example workflow file: .github/workflows/deploy.yaml

---

🐳 Docker Build & Deployment

Build Locally

docker build -t gemini-rag-assistant .
docker run -p 8000:8000 gemini-rag-assistant

Automatic Deployment

After CI tests pass:

• Docker image is built using requirements.txt
• Image is pushed to DockerHub
• Deploy on Render.com with correct API keys injected via Render Dashboard

---

🌐 Live Deployment

direct link : https://rag-assistant-diwc.onrender.com/

---

🧪 Example Chat Session

• Live Conversation

---

🛡️ File Deduplication

Files are hashed using SHA256 to avoid re-uploading and re-indexing the same file with different names.

---

🔂 LangSmith Tracing

• Language logs

---

📦 Future Enhancements

• ✅ Add authentication
• 🔍 Support other formats like DOCX
• 📈 Usage analytics dashboard
• 💬 WebSocket support for live chat streaming

---

📹 Reference Video

• Youtube Video

Rag-Assistant-Video

---

🧑‍💻 Maintainer

• Name: Yash
• GitHub: yashh2417
• LinkedIn: yashh2417
• Email: yashh2417@gmail.com

---

📝 License

MIT License – feel free to use and modify.

---