RAG F1 Bot: retrieval augmented generation with reranking
A Formula 1 question answering bot built on retrieval augmented generation, with AstraDB search, Cohere Rerank and an LLM judge for evaluation.
Problem
A plain language model answers Formula 1 questions from memory and gets facts wrong.
What I built
A RAG pipeline that retrieves passages from AstraDB, reranks them with Cohere Rerank, and has the model answer from them. Quality is measured with an LLM as judge evaluation.
Stack
Python · AstraDB · Cohere Rerank · LLM as judge
Result
Mean reciprocal rank (MRR) of 0.955 on the evaluation set.