Chapter the Second
A short
catalog of practices.
Four areas we work in regularly. Each can be a fixed-scope engagement or an ongoing arrangement, depending on what you need.
— Index —
— Practice 01
Useful AI features.
Designing and shipping AI features end-to-end: picking the right model, building evaluation, and getting the user-facing details right so the result feels useful in day-to-day work — not just a demo.
— 01
Applied research
An honest read on what current models can and can’t do for your specific problem.
— 02
Production AI
RAG, agents, and structured outputs wired into your stack, with the evals to keep them honest.
— 03
AI-native UX
Interface details that keep humans in the loop where judgment matters.
— Practice 02
Boring infrastructure.
Infrastructure designed to be simple to operate: predictable cost, observability that answers real questions, and migrations planned in steps. We tend to pick the unfashionable tool when it fits the job better.
— 01
Architecture
Greenfield design or a fresh look at what you have. We pick boring tech on purpose, and write down why.
— 02
Migrations
Phased, reversible plans rather than big bangs. Lights stay on.
— 03
Observability
Logs, traces, and dashboards aimed at the questions on-call actually asks.
— Practice 03
The data plumbing.
Most ML work succeeds or stalls long before the model. We help with the pipelines, feature stores, evaluation harnesses, and labeling tools underneath. Less exciting than the model, but usually where the leverage is.
— 01
Pipelines
Streaming and batch, with backfills, replays, and lineage that holds up to audits.
— 02
Feature & vector stores
Retrieval that’s fast, affordable, and honest about its tradeoffs.
— 03
Evaluation
Test harnesses that catch regressions before users do.
— Practice 04
A second opinion.
When you want an outside read on architecture, build-vs-buy, or roadmap tradeoffs. Useful before committing to a direction — and we have nothing else to sell you besides the review itself.
— 01
Reviews
Architecture, code, and roadmap reviews delivered as written memos.
— 02
Due diligence
Technical DD for acquisitions and investments — fast and candid.
— 03
Working sessions
Time with engineering leaders working through a specific decision.
— Engagement shape —
Three ways
to work together.
1 — 2 weeks
Diagnostic
A focused review of an existing system or roadmap, delivered as a written memo: what we found, what we’d do, and what we’d skip.
6 — 12 weeks
Build
A scoped engagement to ship something specific — an AI feature, a migration, or a platform piece — with code in your repo and regular check-ins.
Ongoing
Retainer
A monthly arrangement for teams who want a senior set of eyes on call: code reviews, architecture conversations, and the occasional rescue.