Building production LLM systems
Read the sequence on context, retrieval, evidence and abstention.
Reading order
- Beyond prompt engineering: building a reliable context system
- Context engineering: combining prompts, retrieval, memory and tools
- Compressing and ordering context so a model can use the evidence
- Evaluating RAG retrieval: diagnosing recall, ranking and attribution
- Citations and attribution in RAG: tracing claims back to evidence
- RAG ingestion and chunking: turning documents into retrievable evidence
- Hybrid retrieval and reranking in RAG: combining keywords, vectors and a reranker
- Low confidence in RAG: when to answer, clarify or hand off