AI engineering insights.
AI engineering insights on shipping LLM systems to production: evals, RAG, AI agents, cost and the platform work around them. Written by the engineers who ran the code.
Latest AI engineering insights and articles
AI engineering essays, not thought leadership.
Every post starts with a problem we hit shipping AI for a client, like the eval harness behind our prior-auth case study. Want the pattern applied to your product? Talk to our AI agent and LLM development team.
LLM evals & quality
Eval harnesses, labeled test sets, regression gates and red-teaming. How we know a model change is actually an improvement.
RAG & retrieval
Chunking, embeddings, pgvector vs. dedicated vector stores, and when retrieval beats fine-tuning.
AI agents & automation
Tool use, guardrails and human-in-the-loop review. Agent patterns that hold up in production, and the ones that don't.
LLM cost & latency
Cost per request, token budgets, caching, model routing and p95 latency. The numbers that decide whether an AI feature survives.
AI in regulated industries
HIPAA, SOC 2 and PCI DSS in practice: audit logs, data privacy and model governance for healthcare and finance.
Platform & delivery
Architecture trade-offs, feature flags, observability and migrations. The product engineering that makes AI work in production.
Engineers shipping the work, then writing about it.
Every essay is bylined by an engineer who worked on the code it describes. No ghostwriters, no marketing rewrites.
Hamid Khan, Founding partner
Architecture, sales engineering and the founder-engineer view. 11 years of shipped product across FinTech and HealthTech.
Engineering principals
The senior engineers who run engagements, writing on LLM evals, RAG, agents, multi-tenant architecture and delivery.
Domain specialists
HIPAA, SOC 2, PCI DSS and payments rails. The people who take clients through certification, writing about how it works.
Guest engineers
Occasional posts from clients and peers. Different perspectives, same bar for substance.
Things we'd rather not say.
Four kinds of post you won't find here, because they waste your time and ours. Want more depth instead? Request one of our AI whitepapers or see how we work in our AI delivery playbooks.
Listicles & “X AI tools you need”
If a search engine can replace the post, we haven't earned your attention. We write only when we have a take.
Generic AI primers
Model vendors already explain what an LLM is. We write about specific patterns we've shipped to production.
Recycled vendor content
If a cloud or model provider already published it, we don't republish a variant.
AI-generated drafts
We build with AI every day, but we don't use it to write these. The drafts, and the opinions, are ours.
One letter, monthly
What we shipped, what broke and what we learned running AI in production. Once a month, no filler.