RAG Assistant on Your Docs

This demo shows a retrieval-augmented generation (RAG) assistant that answers questions grounded in your own content — documentation, policies, knowledge bases — instead of guessing. Every answer comes with citations, and the whole pipeline is observable.

What it demonstrates

• Grounded answers — responses are built from retrieved passages of your documents, with citations back to the source.
• Semantic search — documents are chunked and embedded into a vector store, then retrieved by meaning, not just keywords.
• Quality guardrails — when the documents don’t contain an answer, the assistant says so rather than hallucinating.
• Full instrumentation — retrieval, embedding, and model calls are traced with OpenTelemetry, so quality and cost are visible.

How it works

1. Your documents are ingested, chunked, and embedded into a vector index.
2. A user question is embedded and matched against the index to retrieve the most relevant passages.
3. Those passages are passed to the model as context, which composes an answer with citations.
4. Each step is traced and measured, so you can tune relevance and watch cost.

Why it matters

RAG is the most reliable way to put AI on top of your knowledge without retraining a model. Done right — with citations and observability — it’s trustworthy enough for real internal and customer-facing use.

This combines our AI integration and observability work — from custom assistants to the dashboards that keep them honest.