Build a RAG that refuses to hallucinate

The brief

Anyone can wire a retriever to a model. This challenge scores whether your system knows when to say it does not know. You get the corpus, an eval set with unanswerable questions planted in it, and a cost ceiling per query. Chunking strategy, reranking and refusal behaviour are where entries separate.

What you submit

• A runnable RAG pipeline with your retrieval configuration documented
• Eval results on the provided question set, including the unanswerable ones
• A short note on cost per query and what you traded away for it

How to participate

1. 01Register free: create your account and tell us a little about yourself.
2. 02Clone the starter repo and download the corpus.
3. 03Build, evaluate, and keep your cost per query under the ceiling.
4. 04Submit the repo link; mentors run your eval script themselves.

How it’s judged

• Refusal accuracy on unanswerable questions
• Retrieval precision
• Cost per query
• Reproducibility

What you walk away with

Feedback for every entry · Top 3 win a full flagship seat

Resources