20 years running operations at scale — from Army reconnaissance data to Walmart's catalog infrastructure. Now I build the AI systems that automate what I spent two decades doing by hand.
// work
Production systems, not demos. Each one runs on a schedule, handles its own errors, and does real work.
A fleet of 40+ autonomous AI agents that runs my businesses daily. A morning brief agent synthesizes overnight signals from 12 sources. Executive agents for marketing, finance, content, security, and engineering each run on their own schedule, log outcomes, and propose their own improvements. When one fails, it self-diagnoses and writes the fix to a proposal queue. The system has been running continuously for 6+ months.
Key design decisions: agents communicate via a shared state bus rather than direct calls — any agent can be added, removed, or paused without breaking the fleet. Each agent writes structured JSON logs so the oracle can audit its own behavior and calibrate its confidence over time.
Production proxy that sits between Claude Code / any OpenAI-compatible client and both local models and Anthropic's API. Routes by model name, request complexity, and token count. Strips PII before any cloud call. Tracks cost per session. Powers the entire oracle fleet — no direct API calls anywhere.
Parses USAA and Zelle SMS alerts in real time → deterministic NLP extracts merchant, amount, and sender → routes to the correct financial ledger (D1 KV or Excel, depending on the account). Handles edge cases: duplicate detection, multi-account routing, and Telegram confirmations. Zero missed transactions across 6+ months.
Affiliate recipe platform with a fully automated content pipeline. A daily exec agent finds trending restaurant dishes, generates copycat recipes, sources affiliate product links, runs image QA, and deploys — all without intervention. The pipeline handles swap image validation, ASIN caching, and dead-link recovery automatically. Infrastructure is production-grade; growth is an organic audience problem, not a technical one.
Personal finance dashboard backed by Plaid. Real-time account balances, 30-day spend/income breakdown, categorized transactions, and a recurring charge detector that diffs the merchant set against a rolling 45-day window. Daily sync via cron. Built on Cloudflare D1 + KV.
Tenant-facing payment portal with landlord admin dashboard. Tracks rent payments from two tenants via dual-ledger (Excel + Cloudflare KV), with Zelle payment detection piped in from the bridge. Landlord admin shows real-time ledger state with payment history.
// capabilities
The full picture of what I bring, built from production systems rather than tutorials.
// background
20+ years building and running systems at scale — from Army reconnaissance data to Walmart's catalog infrastructure to autonomous AI fleets. The domain expertise and the technical work are the same discipline.
// let's work together
20 years of domain expertise in e-commerce, operations, and data — combined with production AI systems I built and run today. Best fit: AI Engineer or ML Engineer at a company where domain knowledge accelerates the technical work. Remote or Philadelphia area.