Skip to content

Service · 03

Frontier AI engineering.

Your AI works in the notebook. Production is a different stack. I run eval infrastructure, prompt-deployment pipelines, and the orchestration between model and user. Engagements are ongoing; specifics live under NDA.

A

Engagement shapes

Four shapes, increasing in commitment. The right shape depends on whether you need an architecture review, a shipped artifact, an ongoing partner, or full ownership of the AI stack.

Advisory

Scope
AI architecture review + monthly LLM strategy pressure-test. Model choice, eval design, failure-mode mitigation.
Duration
4-6 hrs/month, ongoing.
First-month deliverable
AI stack audit + production-readiness assessment. Named gaps with prioritized fix sequence.

Project

Scope
Production AI feature build — RAG, agent, eval pipeline, orchestration layer. Defined-scope shipped artifact.
Duration
6-12 weeks, defined scope.
First-month deliverable
Technical design + eval infrastructure + first production deployment by close. Documentation handed to your team.

Retainer

Scope
Embedded AI engineering partnership. Continuous integration of new model capabilities, ongoing eval work, prompt iteration.
Duration
Month-to-month, 6-month minimum.
First-month deliverable
First eval framework iteration shipped + prompt-deployment pipeline live + documented failure-mode inventory in month one.

Embedded

Scope
Acting head of AI engineering for the engagement window. Owns architecture, deployment, and the bar for production AI quality.
Duration
3-6 months, 3-4 days/week.
First-month deliverable
AI architecture documented + first production deployment shipped + eval infrastructure foundations live + team onboarded by end of month one.

B

What “production-grade” means here

The phrase carries a specific stack: eval infrastructure that fires on every change, continuous deployment of prompts (not just model versions), confidence thresholds and refusal patterns on the retrieval layer, and a documented bar for what ships vs. what waits. Frontier capability that lives in a notebook is not production. The engagement bar is “deployed, observed, iterated.”

If your stack is pre-production and you’re solo, the playbook covers most of what the advisory shape does — for $149, not $5K a month.

Next step

Discovery call for AI engagements.

The call covers your current stack, the failure modes that worry you, and whether the engagement shape on this page matches what you actually need.