Restaurant owner-operator building agentic AI systems for live business operations.
I have owned and operated restaurants since 2013, previously across two locations and now in an approximately $1.4M business. The operation has delivered 11 consecutive years of positive same-store sales and top-20% performance within its franchise system.
I built an AI operating layer for recurring work that could not stay trapped in spreadsheets, inboxes, and manager memory. It monitors live systems, applies business rules, acts within defined authority, verifies the result, and hands exceptions to a person. It supports third-party delivery profitability, recruiting coordination, guest care, catering, reporting, and system health.
- Pausing six consistently unprofitable delivery campaigns stopped an estimated $500-$900 per week in contribution leakage.
- A matched delivery-ad comparison improved ROAS from 3.09x to 4.01x while spend fell 51%.
- Own the operating problem, economics, records, rules, authority limits, failure paths, and acceptance criteria.
- Direct AI-assisted implementation across APIs, browser workflows, scheduled jobs, and reporting systems.
- Test edge cases against production evidence and verify outcomes in the live source system.
- Keep customer promises, sensitive cases, and ambiguous decisions under human control.
My technical range includes Python, JavaScript, Go, SQLite, REST APIs, Playwright, n8n, Docker, and Git. Those tools support the operating system; they are not the headline.
A scheduled contribution-management system across third-party delivery marketplaces and the POS. It pulls campaign, payout, order, daypart, and store data; calculates contribution after commission, food, packaging, discounts, and ad spend; then takes one reversible action or none. Every action is checked against the live source.
A controlled browser workflow over a guest-care inbox. It binds every action to the correct case, drafts only within policy, routes sensitive and ambiguous cases to a person, verifies staff notification before a customer promise, and alarms when expected work goes silent.
A workflow across Indeed and Homebase that normalizes candidate state, preserves durable exclusions, suppresses duplicate outreach, checks manager availability, and prepares contact only when the record supports it. Managers keep every hiring decision.
Evidence-first restaurant portfolio margin control built for OpenAI Build Week. Deterministic Python owns every financial calculation; the model is a bounded narrative provider; humans approve or reject every proposal; SQLite preserves estimates, decisions, and measured outcomes separately.
- Live demo (free-tier host; allow about 30 seconds to wake)
- Devpost submission
- Demo video
An interpretable soccer feature-engineering package built from tracking and event data, with deterministic output and reproducible validation.
- President and Restaurant Owner-Operator, R3M LLC (2013-present)
- Approximately $1.4M annual operating scale with full P&L accountability
- Previously operated two locations simultaneously
- Own labor, inventory, vendors, recruiting, guest experience, catering, third-party delivery, reporting, and operating controls
- Built SOPs for reporting, cash reconciliation, food safety, hiring, training, and day-to-day execution
- ServSafe certified and ServSafe Proctor
- A stop rule needs a path that can satisfy it.
- Escalation must reach a person on a channel they read.
- Silence is a failure state, so expected work has its own alarms.
- A system should never promise a customer something the staff have not been told about.
- Important actions are verified in the source system and recorded for audit.
I am open to remote operations leadership, agentic workflow and implementation roles, AI evaluation, restaurant-operations consulting, and contract work where operator judgment and verifiable execution matter.