Digital innovation zone - AI chatbot
A self-hosted RAG chatbot that answers questions about DIZ EDIH's services for public institutions — eligibility, funding, courses and the innovation ecosystem — grounded strictly in DIZ's own knowledge base, running entirely on their own infrastructure with no per-query API costs.
Digital Innovation Zone (DIZ) EDIH — the European Digital Innovation Hub for Romania's North-East region — needed a way for public institutions (municipalities, hospitals, schools, cultural organizations) to quickly understand which of its free EU-funded services they're eligible for, without digging through eligibility rules, funding structures and course catalogs spread across multiple documents. We built a retrieval-augmented generation (RAG) chatbot that answers in natural language, grounded strictly in DIZ's own knowledge base — service catalogs, the funding-access pathway, course listings, the legal/glossary reference, and an FAQ — so it never invents an answer outside what DIZ actually offers.
The whole stack runs self-hosted on DIZ's own infrastructure rather than calling a third-party LLM API: a NestJS backend orchestrates retrieval against a Qdrant vector store and generation through a locally-run Ollama model, with Redis caching and streamed responses for a responsive chat experience on a React front end. Admins manage the knowledge base directly — uploading or removing PDF, DOCX, TXT and Markdown source documents — so answers stay current as DIZ's programs evolve, with zero per-query API costs and no institutional data leaving their own servers.