I will fix your rag chatbot, ai agent or llm giving wrong or hallucinated answers

I fix one named failure in a RAG or retrieval pipeline you already have.

You name the question that comes back wrong: it invents an answer, misses a document you know is there, cites the wrong source, or answers when it should say it does not know.

What you get

- The fix, as a diff you can read and revert

- Your failing question passing: it fails before the fix and passes after

- What was actually wrong, named: the chunking, the embedding, the top-k, the filter, or the prompt

Standard adds the eval cases that lock it: the question, the source it must cite, and the refuse-when-empty case, so the failure cannot come back quietly.

Premium is for when the wrong answer is a symptom. One named question path, up to 3 failures on it, citations you can open, and a written list of what I left alone.

How it works: send the failing question and the repo. I reproduce it, fix it, and send the diff.

Since 2011 I have run RedRobot, building production data systems on client infrastructure. Before that I ran MySQL and MariaDB in Japan and led DataStax's market entry there.

Existing pipelines only. I do not build a RAG from zero, ingest a new corpus, or train a model.