Project Detail
Searchable RAG Copilot
An enterprise search system that lets users query internal documents and receive cited, grounded answers using Retrieval-Augmented Generation
AIBackendPythonFastAPI
In progressTarget completion August 2026
The Problem
Large organisations store knowledge across hundreds of documents that are difficult to search meaningfully. Keyword search returns results, but not answers. This project addresses that gap.
What I Built
A Python API built with FastAPI that accepts natural language queries, retrieves relevant document chunks using semantic search, and returns cited answers generated by a language model.
Tech Stack
- Azure OpenAI
- Azure AI Search
- Azure Blob Storage
- FastAPI
- LangChain
- Ollama
- Mistral 7B
- ChromaDB
Decisions I Made
- Designed the local-to-Azure swap architecture so only environment variables need to change for deployment.
- Chose ChromaDB for local vector storage because it is simple and supports persistence.
- Implemented simulated RBAC in Python to model enterprise access control without exposing live credentials.
What I Learned
- Chunking strategy, embedding quality, and retrieval tuning matter more than model choice alone.
- Understanding the retrieval pipeline end to end was the main technical takeaway.