Applied RAG · LLM

Language models (LLMs) know a bit of everything, but they don't know your domain or your sources — and sometimes they make things up. Retrieval-augmented generation (RAG) fixes this: before answering, the system retrieves the most relevant fragments from its own knowledge base and grounds the answer in them, citing where each fact came from.

I build these systems for sensitive domains — a migrant storytelling archive, postpartum support for women, food systems — where it matters that the answer is anchored in verifiable sources rather than the model's opaque memory.

The difference, at a glance:

Explore it

The heart of a RAG system is search. In this demo, a small language model runs inside your browser (no server, no API): it understands your question and finds, by its meaning, the fragments that answer it — even when they don't use the same words. Try turning the search off to see why an assistant without sources is so risky.

The archive content is synthetic and illustrative: it demonstrates the mechanism, not a real source of advice.