Fractional AI engineering
Why hire me — fractionally
Most companies do not need another AI demo. They need someone who will embed with their team part-time and put a working, verified AI system into production — on their real data, wired into the tools they already run. That is the only thing I do, and I have already done it for a business I own.
The case in one line
I was the customer before I was the developer
I run a Michigan home-services company, could not buy the software it needed, so I built it — and that is how I became a forward-deployed engineer: someone who ships production systems on real, messy data inside a business that is already running. As a fractional hire you get that operator’s judgment and a shipped track record, at a fraction of a full-time senior AI engineer’s cost.
Why fractional is the right call
One or two systems, not a year of headcount
The math
A full-time senior AI engineer runs $340–470K all-in in year one. Most teams do not have a year of full-time AI work — they have one or two systems that need to actually ship. At 8–12 hrs/week I am roughly $90–144K/yr, and you only pay while there is work worth doing.
The speed
No three-month hiring loop, no ramp. I start on your problem in week one, ship something into production inside the first engagement, and scale hours up or down as the roadmap changes.
Rate figures are market benchmarks for fractional AI engineering (roughly $150–250/hr direct); the full-time comparison is total comp, not salary. Your actual scope sets the number.
What you’re actually getting
Five things, and none of them is a demo
Deployment, not demos
The model is the easy part now. The work is integration — mapping how your team actually operates, wiring into the systems you already use, and earning trust in the data. That middle ground between engineering, product and the customer is where I live.
A reliability-first approach to AI
I build the verification layer first: evaluation harnesses, retrieval-quality gates, agents with scoped tools and controlled state. A bad model output or a changed prompt should not be able to reach your user — an eval gate stops it. That is the difference between "it worked in the demo" and "it is safe in production."
Full-stack ownership on AWS serverless
Python and TypeScript. Lambda, DynamoDB, API Gateway, Cognito, CloudFront, Terraform. I can take a system from data model to deployed and monitored — you do not need to staff around me.
Security by necessity
The data I have shipped on has been PHI, financial, and other people’s customers. Tenant isolation, encryption and least-privilege access are not add-ons; they are how I build.
An operator’s lens
I have run payroll, chased receivables, and felt the cost of software that does not fit the workflow. I build around how your people actually work, not around my schema.
The receipts
Two production systems, built solo, end to end
This is what “shipped” means when I say it.
Live
Home-services operations platform
Owner web and crew mobile on AWS serverless. Email-driven auto-ticketing, a multi-rate billing engine, scheduling, CRM, Plaid and QuickBooks sync, receipt OCR through Claude vision, and a Profit First accounting layer.
Billing, crew payouts and A/R run on roughly 18,000 lines of Python — 51 scheduled automations, 18 of which move real money, every one dry-running unless explicitly told to commit.
See the platform →Built and owned
Telehealth reliability layer
A multi-tenant, HIPAA-ready platform: PostgreSQL row-level security for hard tenant isolation, 608 passing tests, and an LLM reliability layer — an evaluation harness with retrieval-quality gates, so model output is measured and verified before it could reach a patient.
Production-ready. Built and owned by me, with no live tenant — and I will not imply otherwise.
See the platform →The numbers behind the ops platform
The invoiced, collection, transaction and automation figures are actual — from DynamoDB, QuickBooks and the scheduler itself. The ~140 hours is a conservative estimate and is labelled as one. I also run a live agency CRM in AWS. I do not present trading or investing results as a track record.
How a fractional engagement runs
Four moves, then I am optional
Cadence
A fixed block of hours per week, a weekly working session, async the rest. You always know what shipped and what is next.
Engagement shapes
A scoped project with a fixed outcome, an ongoing retainer at 8–20 hrs/week, or a short paid discovery to de-risk the first one. Month-to-month — I keep the engagement by earning it.
The goal is to make myself optional
Documented, tested, owned by your team. Built to run without me is the whole point.
Where I’m the right fit
And where I am not
Hire me if
You are deploying AI where a wrong answer is costly — health, finance, ops, regulated data — and you need it to actually work in production, integrated and verified, not a prototype that impresses in a meeting.
Don’t hire me if
You want a large research team, frontier-model training, or a body to fill a seat. I am one senior builder who ships end to end — that is the value and the limit.
Tell me what you are trying to put into production
A short paid discovery is the cheapest way to find out whether this works. If I am not the right person for it, I will say so — and it costs you one email.