Six-to-twelve-week advisory engagements with engineering teams moving language models from prototype to production. Direct, hands-on, and ruthlessly scoped.
I work with one team at a time, embedded as a senior advisor for the duration. You get me in your standups, your design reviews, your eval triage and your VP-of-Engineering 1:1s. I don't write production code; I make sure yours is the right production code.
Default scopeis twelve weeks, one engagement at a time, two clients at most per quarter. I don't take more. If you need a delivery team, I'll introduce you to two I trust.
// four bets per engagement
Map your system's actual failure surface (per the SAAL framework). Identify which two seams are killing you. Most teams know about the third already and have it covered.
Design the eval set that survives a quarter of drift. Set up the labeling cadence, decide who owns it, and write the playbook for when it disagrees with prod.
When I leave, your platform team should be able to operate this system without me. The handoff is a real document, drafted in week 8, ratified in week 11.
Two hours a week, off the record, with your senior IC. Everything I know about leading an LLM team, transferred.
This isn't an AI strategy retreat. There are no slides at the end.
This isn't a team-augmentation contract. I'm one person; the work is advisory.
This isn't a sales call. If you want to know whether you should be using AI at all, the answer is in essay 015; we'll figure out the rest on the intro call.
// open work in progress
Retrieval-Augmented Generation done right means more than wiring a vector store to an LLM. The RAG Axis covers the ingestion pipeline, chunking strategies, retrieval quality, and the failure modes that only show up in production.
A common space for all communities part of the AI revolution. Cespace hosts builders, learners, and thinkers who are shipping production AI systems.
Intro calls are 30 minutes. I respond to thoughtful emails within a week. Vague ones, sometime later.