Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
RAG_retriever: Secure Local RAG
This talk demonstrates a local RAG pipeline that retrieves, extracts, classifies, and verifies information from documents, ensuring grounded results and filtering out misleading content.
RAG_retriever is a local document intelligence pipeline that retrieves information from your files and then verifies its own output before returning results. Each stage (retrieve, extract, classify, verify) runs as a separate LLM call with a narrow contract and a pass/fail gate. The verification stage checks whether extracted claims actually appear in the source chunks and rejects anything it can’t ground. In the demo I’ll run the full pipeline live on a document set that includes embedded prompt injections and misleading content, walk through the execution trace so the audience can see how each micro-AI stage processes and filters information, and present benchmark results on retrieval accuracy. Everything runs locally on Ollama with no cloud dependencies.
Hybrid RAG pipeline using vector/BM25 retrieval with iterative LLM verification.