Generative AIDemo not connected
Advanced RAG Knowledge Assistant
A document question-answering assistant with semantic chunking, reranking, query rewriting and a validated FastAPI layer.
- Python
- LangChain
- ChromaDB
- Embeddings
- FastAPI
- Pydantic
- Gradio
Problem
Answering questions over a large body of unstructured documents requires retrieving the right passages reliably before a language model can respond usefully.
Approach
Built a preprocessing and chunking pipeline with LangChain, stored embeddings in ChromaDB, and connected a retriever to an LLM behind a Gradio chat interface. A custom retrieval layer was then built from scratch to improve result quality.
Key features
- 01Document preprocessing and semantic chunking
- 02Embedding storage and similarity retrieval with ChromaDB
- 03Reranking and query expansion / rewriting
- 04Embedding-space visualisation in 2D and 3D to compare embedding models
- 05Multi-turn conversation history in a Gradio UI
- 06FastAPI endpoints with API-key authentication and Pydantic request / response validation
Demo
A working deployment hasn't been connected for this project yet. It will appear here once a hosted link is added.
Next project
Multimodal AI Assistant Platform