Gemini-Powered RAG Chatbot using LangChain

This project demonstrates a Retrieval-Augmented Generation (RAG) pipeline using:

• Google's Gemini 1.5 Flash (for chat and embeddings),
• FAISS (for vector search),
• LangChain (for orchestration),
• and LangSmith (for tracing/debugging).

---

Project Overview

The RAG pipeline has two main components:

1. Indexing Pipeline

Used to preprocess and store documents in a searchable format.

Steps:

Load --> Split --> Embed --> Store

• Load: Load documents using Document loader
• Split: (optional) Use text splitters for large texts
• Embed: Use Gemini embeddings (embedding-001)
• Store: Store vectors in FAISS for fast retrieval

---

2. Retrieval + Generation

Used at inference time to fetch relevant content and generate answers.

Flow:

Question --> Retrieve --> Prompt --> LLM --> Answer

• Retrieve: Pull top-k relevant chunks from FAISS
• Prompt: Combine question + retrieved content
• LLM: Use Gemini 1.5 Flash to generate the answer
• Answer: Return final response to user

---

How to Run

Prerequisites

Install dependencies:

pip install -r requirements.txt

.env File

Create a .env file in the root directory:

GOOGLE_API_KEY=your_google_api_key
LANGCHAIN_API_KEY=your_langsmith_api_key
LANGCHAIN_PROJECT=your_project_name   # optional

---

Run the App

python rag_app.py

You'll be prompted to ask a question, and the system will return a generated answer along with the retrieved source documents.

---

FAISS Index

The FAISS vector index is saved locally as:

my_faiss_index/

You can reuse it later without re-indexing documents.

---

LangSmith Integration

The code includes optional LangSmith support:

• Enables tracing and monitoring your LLM pipeline.
• Helps debug prompt flow and understand model behavior.

📌 Setup your LangSmith account to get the LANGCHAIN_API_KEY.

---

Google Gemini API Billing

Be cautious about API usage:

• Google offers free tier via AI Studio.
• Use budget alerts in your Google Cloud Console.
• Monitor usage and avoid unnecessary charges.

🔗 Gemini Pricing

---

Tech Stack

Tool	Purpose
Gemini Flash	Fast LLM for generation
FAISS	Vector store for retrieval
LangChain	Orchestrates RAG pipeline
LangSmith	Logs/traces LLM executions
dotenv	Loads API keys from .env

---

Future Improvements

• Add PDF/Text file loaders
• Integrate a UI (e.g., Streamlit or Gradio)
• Add evaluation metrics with LangSmith
• Implement text chunking and metadata

---

Example Query

Ask a question: What is RAG?

 Answer: RAG stands for Retrieval-Augmented Generation...

Sources:
[1] RAG stands for Retrieval-Augmented Generation.
[2] LangChain is a framework...

---

References

https://python.langchain.com/docs/tutorials/rag/