AI
Shipping LLM features that survive contact with real users: evals, agents, retrieval and restraint.
Golden evals: mining support tickets for test sets
Your support inbox is the best LLM eval dataset you'll ever get. How to mine tickets into versioned golden sets that catch regressions before users do.
Evals before features: how we measure LLM quality
How we build evaluation harnesses for LLM features: golden datasets, rubric graders, human review lanes and regression tracking that stakeholders can read.
Streaming UX: making latency feel like thought
Token streaming is a design material, not a loading state. Skeleton-to-stream choreography, citation timing, honest cancellation, and perceived-performance numbers.
Prompt libraries are a design system too
Prompts scattered through a codebase rot like untracked CSS. How we build prompt libraries: structure, versioning, test coverage, and a shared vocabulary.
RAG pitfalls we hit so you don't have to
Retrieval in production humbles everyone. The chunking trade-offs, citation bugs, stale knowledge and eval failures that taught us how RAG actually breaks.
Keeping AI output on-brand: voice guardrails that hold
Brand voice survives generation only if it's engineered: voice charts as prompts, example banks, style evals, review lanes, and a graceful drift fallback.
Cost engineering for LLM features
Unit economics for AI features: model routing, caching layers, context trimming, batch APIs, and dashboards that translate tokens into dollars per user.