FOR FOUNDERS · FRACTIONAL AI CTO
Fractional AI CTO for founders staking the company on an AI bet. Strategy pressure-testing, architecture review, and de-risking the build — most engagements have production artifacts within 90 days.
The real bottleneck
You can ship faster with AI tools. The bottleneck that doesn't compress is figuring out whether what you're building is the right thing — you can go in the wrong direction faster too.
Your demo is great. Investors love it. But you don't know how it holds at production volume, your evals are green while customers are unhappy, and every time a new model drops you're re-litigating every architecture decision.
Token costs are eating your margin. The dev shop delivered a beautiful demo with no evals and no production readiness. Your RAG pipeline works in the notebook and breaks on real queries. You're burning runway on the wrong bet.
You can't hire a VP of AI at $500K. You need senior judgment in a smaller package — someone who's made these calls before and can say “I've seen this pattern, here's how it plays out.”
What I do with founders
Your strategy is a hypothesis. The question is whether it holds when GPT-5 drops, when your token costs hit production volume, or when your first real customer sends the edge case your eval set never saw. We run it hard before you stake the company on it.
The choices you make in the next 90 days — model, RAG pipeline, agent loop structure, fine-tune vs. prompt — are the ones you'll be living with in month 18. I've seen which bets hold and which don't. We pressure-test before you lock in.
Demos lie. A 95% eval score can mask the exact failure mode your angriest customers will find. I help you build the eval infrastructure that catches real failures, model the inference cost at production volume before it surprises you, and ship without painting yourself into a corner.
From the field
Series B fintech · Q4 2025
Problem
Inference costs tripled month-over-month with no traceability. Board was asking CFO questions the team couldn't answer.
Outcome
Inference cost reduced 58% while throughput increased. First time the team could answer a board question about cost-per-transaction.
Consumer marketplace · Q1 2026
Problem
Two years of AI feature work, one agent in production, zero eval framework. Releases were manual spot-checks by a PM with a spreadsheet.
Outcome
Eval cycle dropped from 2 weeks manual to 3 hours automated. Next model upgrade shipped in one sprint instead of a quarter.
Early-stage AI startup · Q3 2025
Problem
Technical founder making irreversible architecture choices alone. No senior AI product peer to pressure-test the RAG pipeline or model selection before Series A.
Outcome
Identified the model lock-in risk before Series A close. Rebuilt eval strategy from production edge cases, not demo assumptions. Raised on a credible technical narrative.
Selective by necessity — I work with a few teams at a time. We figure out together if there's real fit — no sales process.