LLM-powered natural language interface for internal company data
Screenshots
The problem was simple: staff were spending too much time searching through internal documents, databases and spreadsheets for answers that should have been instant.
I architected a RAG (Retrieval-Augmented Generation) pipeline using Ollama running LLaMA locally, LangChain for orchestration, and a FastAPI backend exposing a clean API. The frontend is a React interface that feels like a chat app but queries structured internal data.
The entire system runs on-premises using Docker containers managed with Dokploy, ensuring zero data leaves the company network — a critical requirement for internal tooling. Documents are indexed into a vector store and retrieved semantically before being passed to the LLM for synthesis.
This was a self-initiated project: I identified the productivity gap, proposed the architecture, and delivered it end-to-end.