RAG Knowledge Assistant

Python + LangChain + OpenAI + FAISS + FastAPI

Overview

A simple Retrieval-Augmented Generation (RAG) assistant that answers questions over your own documents. It loads local text files, indexes them in a vector database (FAISS), and uses an LLM to generate answers with citations like [DOC_1].

Problem

Traditional keyword search is weak at answering natural language questions over internal documents.
This project shows how to build a small end‑to‑end RAG pipeline that can:

• Ingest your documents.
• Retrieve the most relevant chunks.
• Generate grounded answers that point back to the source text.

Tech Stack

• Python
• LangChain
• OpenAI API (embeddings + chat)
• FAISS vector store
• FastAPI
• CLI helper script

Architecture

1. Ingestion
   src/ingest.py loads .txt files from src/data/, splits them into chunks, creates embeddings, and saves a FAISS index on disk.

2. Retrieval + Generation
   src/rag_chain.py loads the FAISS index, retrieves the top‑k chunks for a user query, and calls an LLM with the retrieved context to generate an answer with simple [DOC_i] citations.

3. Interfaces

   • CLI: src/cli.py → python src/cli.py "your question"
   • API: src/api.py → FastAPI endpoint POST /ask that returns {"answer": ..., "sources": [...]}.

Setup

Prerequisites

• Python 3.10+
• OpenAI API key
• Git

Installation

git clone https://github.com/kunalg06/rag-knowledge-assistant.git cd rag-knowledge-assistant

python -m venv .venv

Windows: .venv\Scripts\Activate.ps1 macOS / Linux: source .venv/bin/activate pip install -r requirements.txt

Configuration

Create a .env file in the project root: OPENAI_API_KEY=your_real_key_here

Do not commit .env to Git.

Usage

1. Add documents

Place your .txt files inside src/data/, for example:

• src/data/doc1.txt
• src/data/doc2.txt
• ...

2. Build the FAISS index

python src/ingest.py This loads the files, splits them into chunks, creates embeddings, and saves the index under faiss_index/.

3. Ask questions via CLI

python src/cli.py "What are these documents about?" You will see:

• The generated answer.
• A list of DOC_i: filename sources that were used.

4. Run the FastAPI server

uvicorn src.api:app --reload Then send a request: curl -X POST "http://127.0.0.1:8000/ask" -H "Content-Type: application/json" -d "{"query": "What is inside these documents?"}"

The response contains answer and sources in JSON.

Evaluation

Basic evaluation is implemented in src/eval.py:

• data/eval.jsonl holds small evaluation examples:
  {"question": "...", "expected_keywords": ["..."]}
• The script calls the RAG pipeline for each question and computes very simple keyword‑based precision/recall.

Run:

python src/eval.py This prints answers and aggregated metrics.

Future Work

• Support PDFs / DOCX using LangChain document loaders.
• Add a simple web UI (Streamlit or frontend for FastAPI).
• Better evaluation (RAGAS‑style metrics, more test questions).
• Add other embedding providers or vector databases (e.g. Pinecone).