Distribution Management
Senior Software Engineer
Insurance distribution platform for producer onboarding, licensing, and compensation across enterprise carriers.
Service
ML Strategy & Audit
A straight answer on whether the AI thing is worth building.
Sometimes the most valuable deliverable is "don't build this". A short, focused review of your architecture, model choices, and risk surface — ending in a written recommendation you can hand to a board or an engineering team.
Sound familiar?
You probably need this if…
If two or more of these land, this is the right conversation to have.
Leadership wants an AI roadmap and every estimate in it is a guess.
You're choosing between vendors and all the decks say the same three things.
A prototype impressed everyone, and nobody knows what productionising it costs.
Compliance asked what happens to the data, and the answer was vague.
Deliverables
What actually gets built.
Artifacts, not slides — all of it yours to keep and run without me.
- 01
A written architecture review
What you have, what it will and won't survive, and the specific changes that matter — ranked by effort against impact.
- 02
A model and vendor recommendation
Compared on your actual workload rather than public benchmarks: quality, latency, cost at your volume, and how painful switching later would be.
- 03
A risk register
Data handling, prompt injection, PII exposure, hallucination blast radius, vendor lock-in — each one paired with a mitigation, not just a warning.
- 04
A costed roadmap
A phased plan with realistic timelines, the team shape each phase needs, and explicit kill criteria for the parts that might not work.
- 05
A readout with your team
I walk engineers and stakeholders through the findings and answer the hard questions live, rather than emailing a PDF and disappearing.
Process
How this one runs.
Three phases, with something demoable at the end of every week.
- 01
Read everything
Code, docs, dashboards, and the prototype — plus interviews with the people who have to live with the result.
- 02
Pressure-test
Run the system against realistic inputs and edge cases, and cost the options out at your actual volume.
- 03
Write it down
One document your engineers and your finance team can both read, followed by a live readout.
The stack
What I reach for.
Defaults, not dogma — I'll work in yours where it makes more sense.
- Python
- OpenAI
- Anthropic
- AWS
- Google Cloud
- PostgreSQL
Proof
Where this has shipped.
Real products, in production, with real users on them.
FAQ
Before you ask.
How long does an audit take?
One to three weeks, depending on how much system there is to read. The output is the same either way: a document and a live readout.
What if the answer is "don't build it"?
Then a couple of weeks just saved a quarter of engineering time. That has been the right answer more than once.
Does the review lead into building it?
Often, yes — a review is usually the cheapest way to find out whether the build is worth doing. The document stands on its own either way if you'd rather your own team ran with it.
Related
Often paired with.
Need help with ML Strategy & Audit?
Tell me what you're working on. I'll tell you what it takes — and whether it's worth building at all.