LangChain RAG Q&A Application

This repository contains a robust question-answering (Q&A) application built using the LangChain library. The application retrieves information from web URLs and local documents to provide accurate, cited answers to user queries.

🚀 Recent Enhancements

• Hybrid Search: Combines BM25 (keyword) and Vector search using Reciprocal Rank Fusion (RRF).
• Metadata-Aware Re-ranking: Improved re-ranking that considers document titles/filenames, ensuring better retrieval for author-specific queries.
• Query Expansion: Automatically generates multiple query variations to improve retrieval recall.
• Streamlit Web UI: A modern, interactive web interface for a premium chat experience.
• Improved CLI Commands: Added /sources to quickly list all indexed documents.
• Multi-Source Loading: Native support for PDF, DOCX, and CSV files in addition to Text and Markdown.
• Recursive Loading: Support for recursive directory searching in the docs/ folder.

Features

Core Retrieval Pipeline

• Hybrid Retrieval: Pairs traditional keyword matching (BM25) with semantic vector search.
• Intelligent Re-ranking: Uses a Cross-Encoder model (ms-marco-MiniLM-L-6-v2) to score document relevance, now with metadata awareness.
• Multi-Query Expansion: Uses the LLM to understand different facets of the user's intent.
• Data Persistence: Uses Milvus Lite to store document embeddings locally (milvus_demo.db).

Multi-Source Indexing

• Web URLs: Loads all URLs listed in sources.txt.
• Local Documents: Supports .txt, .md, .pdf, .docx, and .csv files in the docs/ directory, with recursive searching enabled.

User Experience

• Streamlit Web UI: Modern glassmorphism UI with configuration sidebar.
• Interactive CLI: Multi-turn conversations with history and source management.
• Automated Summarization: Transparently manages long conversation context.

---

🖥️ User Interfaces

1. Web UI (Streamlit)

For a premium, interactive experience, use the Streamlit interface:

uv run streamlit run app.py

2. Interactive CLI

For terminal-based usage:

uv run --env-file .env -- rag.py --interactive

Interactive CLI Commands:

• /sources - List all unique files currently in the index.
• /history - Display conversation history.
• /save - Save current conversation to a JSON file.
• /load - Load a previous conversation.
• /clear - Clear current history.
• /exit - Exit the application.

---

Configuration

Quick Start

1. API Key: Add your OpenAI API key to .env: OPENAI_API_KEY=sk-...
2. Sources: Add URLs to sources.txt or files to docs/.
3. Run: uv run --env-file .env -- rag.py

Advanced Configuration (config.yaml)

Key areas in config.yaml:

• retrieval: Adjust k and hybrid search weights.
• document_processing: Add new file patterns to supported_formats and ensure local_docs_dir is correct.
• prompts: Customize the agent's personality and instructions.

---

Project Structure

rag-langchain/
├── rag.py                      # Core RAG logic & CLI
├── app.py                      # Streamlit Web Interface
├── evaluate.py                 # RAGAS evaluation script
├── config.yaml                 # Main configuration
├── sources.txt                 # Web URLs to index
├── docs/                       # Local documents (PDF, DOCX, CSV, MD, TXT)
├── conversations/              # Saved chat histories
└── milvus_demo.db              # Vector database

📊 Evaluation

The project includes an evaluation framework using RAGAS to measure faithfulness, answer relevance, and context precision.

To run the evaluation:

uv run --env-file .env -- evaluate.py

Results will be saved to evaluation_results.csv, providing a quantitative baseline for your RAG performance.

Troubleshooting

Q: Installation fails with onnxruntime? A: Ensure you are using the provided pyproject.toml which pins onnxruntime<1.24.0 for Python 3.10 compatibility.

Q: Retrieval isn't finding new files? A: Run with --force-refresh to rebuild the index (vector store + BM25). This is required any time you add new local documents or URLs.

uv run --env-file .env -- rag.py --force-refresh