Why “faithfulness ≥ 0.90” should gate your deploys
A practical look at turning eval scores into a CI gate — and what to do when a release fails it.
EvalCI/CDRAG
Read articleArticles
Notes from production
Field notes on shipping AI you can trust — evals, RAG, agents, and the engineering discipline around them.
Hybrid retrieval + reranking, explained
BM25 vs dense vs hybrid, and why a reranker is usually the cheapest accuracy you can buy.
RAGRetrievalReranking
Red-teaming an LLM app before users do
Jailbreaks, prompt injection and PII leakage — a checklist mapped to the OWASP LLM Top 10.
SecurityRed-teamOWASP
Agentic RAG without the runaway costs
Tool-use orchestration with guardrails on every hop, and how to keep the bill predictable.
AgentsCost controlRAG
Self-correcting answers with CRAG & Self-RAG
When to retry retrieval, when to abstain, and how to wire the feedback loop.
RAGCRAGSelf-RAG
Observability for LLM systems that actually helps
The traces, metrics and cost signals worth wiring from day one — not after the incident.
PlatformObservabilityOps