Flagship Project

RAGmind.

An agentic retrieval system that answers cybersecurity questions from real reference material, judges whether it found enough to answer, retries when it has not, and cites every source. Built, evaluated, and deployed end to end.

Agentic RAG Hybrid Search Self-Correcting Cited Answers Evaluated

What makes it different

Most question-answering demos retrieve some text, paste it into a model, and hope it was relevant. RAGmind adds the judgment a real system needs:

  1. Hybrid search. It blends semantic meaning with exact keyword matching, so it catches both paraphrased questions and precise terms like specific vulnerability identifiers.
  2. It grades its own retrieval. Before answering, a step checks whether the retrieved material actually covers the question. If not, it rewrites the query and searches again.
  3. It cites and refuses. Every claim points back to a source passage, and if the answer is not in the documents, it says so instead of guessing.
  4. It is measured. A built-in evaluation suite scores retrieval accuracy, answer quality, and how often it correctly refuses questions it cannot answer.

Try it live below. Two ways to use it: ask about the built-in cybersecurity material (try "What is Kerberoasting and how do I mitigate it?"), or use the sidebar to upload your own document (PDF, Word, or text) and ask questions about it. The first question takes a moment to warm up.

Heads up: if the demo shows a sleep screen, just click "Yes, get this app back up" and give it about 30 seconds. The app rests on free hosting when idle and wakes right back up.

The live app works best on a larger screen. Tap below to open it in a new tab.

↗ Open RAGmind
← Back to all work Get in touch