Work

AI that shipped, and kept working.

A track record of AI products built for real users under real constraints: regulated healthcare data, enterprise scale and business cases that had to pay off.

Featured build · My own product

The Homeschool Portfolio

thehomeschoolportfolio.com ↗ A child's completed math worksheet ready to be captured into The Homeschool Portfolio.

The problem

Florida law requires home education families to keep a portfolio of each child's work and have it evaluated every year. In practice that means a year of worksheets in a box, and a stressful few weeks of sorting, labeling and matching work to state standards before the evaluation.

What I built

An AI-powered web app that lets parents build the portfolio as they go, one photo at a time. I designed, built and launched it myself, from product strategy through production operations.

Photo, PDF and email capture: parents snap a picture or forward an email, and the entry is created for them
AI vision reads the work and suggests the subject, a summary and a mastery level
Automatic matching against thousands of Florida education standards using vector search
Google Classroom import and quick bulk logging for busy days
A paginated, evaluator-ready PDF portfolio and a secure share link, generated on demand
Subscription billing, usage metering and automated email notifications

Why it matters for your business

A year-end scramble to organize paperwork against a set of rules isn't unique to homeschooling. It's job files at closeout, receipts at tax time and inspection documents before a walkthrough. The same pattern works anywhere: capture it when it happens, let AI do the reading and organizing, and produce the finished document on demand.

It also shows what's possible now. With modern AI tools, one experienced builder shipped a product that would once have needed a funded team.

Built with Next.js, Supabase (Postgres, pgvector), Anthropic Claude, Stripe and Netlify.

Recent builds

More things I've designed and built

AI where it earns its keep, and dependable automation where that's the better answer.

Agentic AI

Cypress Run Dinner Club

A private web app that runs a recurring group dinner club end to end. Members vote on dates, nominate and vote on restaurants, RSVP and rate past dinners, with automated email notifications at every stage. The host can also skip straight to a reservation when the group already knows where it's going.

How it uses AI

An AI restaurant researcher built on Claude that members chat with about cuisine, vibe, price and location
Live web search plus a custom tool that queries the Google Places API, so recommendations come back as verified restaurant cards with rating, price level, address and contact details
One click adds any suggestion straight to the club's ballot

A small, concrete example of agentic AI: a model that searches the web, calls a real API and hands back grounded, actionable results instead of a plain text answer. The same pattern powers AI assistants that look up real prices, inventory or availability for your customers.

Built with: Next.js, Supabase, Anthropic Claude (tool use and web search), Google Places API, Postmark, Netlify

Operations automation

Lothian Kilt Rentals

An inventory and rental management system for a kilt rental business that ships formal wear to weddings, competitions and events. It replaced a legacy admin tool and brought customers, events and inventory into one place.

What it does

Tracks kilts and accessories by category, size and status, with customer and event records alongside
Assign Rentals, the core workflow: pick a customer and event, set ship and return dates, and see only the items that are actually free for that window
Real-time conflict detection that makes double-booking an item impossible
Reports on what's currently out, upcoming shipments and utilization by category

Not every problem needs AI. Here the right answer was dependable automation and clean, structured data. That foundation is exactly what makes future AI, such as demand forecasting or size substitution suggestions, possible and trustworthy.

Built with: Next.js, Supabase (Postgres), Netlify

Enterprise track record

Healthcare and benefits AI, at scale

Highlights from 26 years inside national healthcare and benefits technology companies. Client and employer details are confidential, so these are described by what was built and why it worked.

Document AI

Automated document verification

Designed, built and launched a generative AI process that reads and checks the supporting documents people submit, such as certificates, statements and records. A review that was entirely manual and analyst-heavy became a mostly automated workflow, with people focused on the exceptions.

Built with: Large language models on AWS Bedrock, forms automation, human-in-the-loop review

For a smaller business

Any business that processes forms by hand: invoices, permits, applications, lien waivers, compliance paperwork.

Data extraction

Health plan document ingestion

Built an AI tool that reads dense health plan documents and converts them into structured data that systems can use directly. It removed most of the manual setup work, reduced errors and omissions, and made a customer-facing decision-support chatbot far easier to keep accurate.

Built with: LLM document parsing, structured extraction, validation rules

For a smaller business

Spec sheets, bids, submittals, vendor price lists and contracts turned into clean, searchable data.

Personalization

AI-driven next-best-action engine

Owned a multi-tenant decision engine that chose the right message for each person across web, email, text, mobile app and call center, serving many employer clients at once. It connected people with benefits and resources at the moments that mattered, and lifted engagement across the board.

Built with: Real-time decisioning, predictive models, omnichannel journeys

For a smaller business

The right follow-up to the right customer at the right time, whether it's a quote, a maintenance reminder or a review request.

Conversational AI

Smarter virtual assistants and call centers

Used natural language processing on large volumes of customer conversations to improve virtual assistants and live support. Real-time transcripts and automatically generated answers helped resolve more questions on the first try and shortened calls.

Built with: NLP, speech-to-text, question-and-answer generation

For a smaller business

AI phone and web intake that answers common questions, books appointments and never misses an after-hours lead.

Anomaly detection

Catching errors in payroll data

Led machine learning work that scanned payroll data for anomalies, surfacing costly errors that people had missed and significantly reducing the manual review effort behind them.

Built with: Machine learning anomaly detection, data quality pipelines

For a smaller business

Flagging billing mistakes, duplicate invoices, job cost overruns and pricing errors before they cost you.

Engagement

Personalized communication journeys

Launched automated, personalized communication journeys for key life and work moments such as onboarding, enrollment, new parenthood and wellbeing, delivered consistently across every channel.

Built with: Marketing automation, segmentation, lifecycle messaging

For a smaller business

Automated customer onboarding, seasonal service campaigns and check-ins that feel personal, not like mass email.

Today I lead product for a health benefits technology company, applying AI to document processing in health plan marketing. Consulting lets me bring that front-line experience to businesses that would never have access to it otherwise.

Have a process that feels like it should be automated?

It probably can be. Tell me about it, and I'll give you a straight answer on what AI can and can't do for it.

Start a conversation