I combine over 25 years of restaurant industry experience with modern AI-assisted development to turn complex business problems into practical, working software. Using tools including Claude Code and Codex, I design, develop, test, and deploy applications that help businesses operate more efficiently.
Loyalty and marketing for local restaurants: customers join on a counter tablet and earn rewards, and Claude-powered agents plan and write the campaigns.
In-store sign-up kiosk, no customer app
Points and check-in rewards, redeemed at the counter
Email, SMS and a shared marketing calendar
AI marketing agents that suggest, write and analyze
Offline-first inventory and ordering for barge and towboat galleys, built by two AI coding agents working in one codebase under a written coordination protocol.
Inventory, PAR levels and purchasing
Chef requests, captain approves, purchase order goes out
Works offline on a shared galley tablet
Two coding agents coordinated with a baton protocol
An agent will write a thousand lines on request. Getting software that does the right thing, and keeps doing it, takes everything else.
Understanding the business
Knowing that an allergy question isn't a support ticket, or that a barge galley's "vendor" is a shopper with a receipt, not a warehouse with a price list.
Breaking the problem down
Turning "we need a CRM" into data models, approval rules and jobs an agent can build one piece at a time.
Managing the agents
Handoff documents, dated decision logs, and one writer at a time, so parallel agents don't overwrite each other.
Evaluating the output
Reading what was built against what was asked for. Agent code that looks right and passes a glance is where the expensive bugs are.
Debugging production
When a job runs five hours early or a page 500s with an empty log, finding the cause in the running system.
Delivering reliably
Testing, deploying, running it daily and owning the fix. Two of these systems have run real businesses day to day.
TypeScript · Claude API · LLM-as-judge · public repo
Complete as an evaluation
A shop's chat assistant handles the easy questions well. What matters is the ones it shouldn't answer: allergies, refunds, angry customers. I wrote four prompt versions and tested each against 40 development cases: 27/40 → 40/40. On nine fresh questions written from real customer messages, the final prompt passed 4/9. That gap is why it isn't called customer-ready.
A library of food-cost, waste and par-level guides, calculators and Excel worksheets for restaurant owners, produced by an automated publishing pipeline. When the first version read like generic AI writing, I rebuilt the pipeline instead of editing the output.
For restaurant software teams
Restaurant industry experience meets AI-powered software development
POS, ordering, loyalty, inventory and AI assistants, built by someone who has worked the line, done the ordering and signed the payroll.