
AliBot
A RAG chatbot for this site, running as a TypeScript Cloudflare Worker. It combines vector search with BM25, reranks the results with a cross-encoder, and has Llama 3.3 70B answer only from the retrieved sources, with citations. A 33-question eval set measures retrieval recall, citation accuracy, and prompt-injection resistance: 100% recall@4 at 1.9 s median latency.















