Bangla QnA Bot (RAG + LangChain)
Overview
Bangla QnA Bot is a Bengali question‑answering system built with Retrieval‑Augmented Generation (RAG). It indexes a document corpus, retrieves the most relevant passages for a query, and then generates grounded answers in Bangla via an LLM orchestrated with LangChain.
Key Features
- End‑to‑end RAG pipeline: loaders → chunking → embeddings → vector store → retriever → generator.
- Bengali‑aware preprocessing for cleaner sentence boundaries and better retrieval.
- Configurable backends: swap embedding models and LLMs without changing app logic.
- Deterministic answering: includes retrieved contexts to keep responses faithful to source.
- CLI / API ready: simple interfaces for local testing or app integration.
Tech Stack
- LangChain for chains, retrievers, and prompt orchestration
- Vector store: FAISS / Chroma (configurable)
- Embeddings: SBERT / Hugging Face models suited to Bangla (configurable)
- LLM: pluggable (OpenAI/Gemini/Local) via LangChain wrappers
- Python, FastAPI/Streamlit (optional UI)
How It Works
- Ingest & Chunk: Documents are loaded and split using Bengali‑aware rules (
।as a primary separator). - Embed & Index: Chunks are embedded and stored in a vector database.
- Retrieve: For a user query, the retriever pulls top‑k semantically similar chunks.
- Generate: The LLM receives the query + retrieved context and returns a Bangla answer.
- (Optional) Cite: Return source snippets for transparency.