Value engineer and agentic AI enablement. I build open-source tooling for AI coding agents, plus native iOS and data products — from App Store releases to live market systems.
App Store release · Portfolio · Public source · GitHub achievements
Small, standalone tools. All MIT, all zero-dependency, all with CI, tagged releases, and test suites — 1,055 tests across the five — and each one checked against a real workstation, not only against fixtures.
| Repo | What it does | Release |
|---|---|---|
| agentsmith | Mines your repo's actual conventions into AGENTS.md — and CLAUDE.md, Cursor rules, and Copilot instructions — with evidence for every rule, then catches drift in CI. No LLM, no network. |
v0.2.1 · 216 tests |
| burnrate | What Claude Code and Codex sessions actually cost, from local logs, plus a hook that caps spend mid-session. Prices cache reads per model and counts each response once — on a real machine, flat cache pricing overstated Claude spend by 50%, and naive counting doubled Codex tokens. | v0.2.0 · 204 tests |
| tripwire | Offline audit of every skill, MCP server, hook, and permission across Claude Code, Claude Desktop, Cursor, and Codex. On a real workstation, v0.1 raised 32 high-severity findings that were all false; v0.2 raises one, and it is real. | v0.2.0 · 196 tests |
| contexttest | A/B testing for instruction files: same task, same commit, two rule sets, paired statistics. Ablates sections, pools repeated runs, compares agents — and checks that the agent actually received the instructions. | v0.3.1 · 113 tests |
| gtm-skills | Nine go-to-market skills for agents — business cases, MEDDPICC deal qualification, pipeline forecasts, pricing, market sizing — on a tested arithmetic engine with an assumption ledger. | v0.2.0 · 326 tests |
| contexttest-findings | Pre-registered experiments on instruction-file rules across Claude Code and Codex. A rule carrying a convention the code lacks moved success from 1/10 to 10/10; a popular scope rule did almost nothing. Includes a published correction of its own first result. | 70 live trials |
A theme runs through all of them: be honest about what you don't know. The business cases grade their own evidence, the convention miner refuses to assert a rule from four files, the cost tool names models it can't price instead of costing them at zero, the security scanner treats a false positive as the failure mode that matters — and when the findings repo discovered its first result was invalid, it published the retraction next to the data.
| Product | State | Built with | Proof |
|---|---|---|---|
| Nalee — honest product-toxin scoring backed by a 2.4M+ product library | Live on the App Store | Expo · Supabase | App Store · Site |
| Lore — geospatial place stories hiding in the streets around you | Live on the App Store | Swift · Supabase | App Store · Source |
| Goals — a cinematic goal-discovery and execution command center for turning intention into daily progress | iOS / TestFlight lane + live web | SwiftUI · TypeScript · Supabase | Live · Private source |
| Tapt — beer discovery, Passport collecting, and a live beer market | Native iOS release lane | Swift · Supabase | Source |
| Mend — a private app that helps couples connect, grow, and stay in sync | TestFlight | Expo · TypeScript · Supabase | Source |
| Precision Algorithms — a published prediction-market models desk | Live web product | Data · Web | Live |
| SqueezeRadar — short-squeeze signals with live price overlays | Offline · redeploy pending | Next.js · Market data | Private source |
| Penny Catcher — volume and flow radar for quiet, low-priced tickers | Offline · redeploy pending | Next.js · Market data | Private source |
agentsmith CI (Python 3.9–3.13 on Linux, macOS, Windows + self-audit) · tripwire CI (plants live attacks, Claude Code and Codex, every run) · contexttest CI (Node 20/22/24 + GitHub Action end to end) · Lore CI and TestFlight · Tapt release automation · Merged pull requests
Value engineering and agentic AI enablement — helping teams adopt coding agents and AI workflows, and measure what they actually return. Previously at Ivanti, where I built the Capability & Maturity Assessment framework.
- Native: Swift, SwiftUI, Expo, React Native
- Web: TypeScript, Next.js, React, Tailwind CSS, Node.js
- Data and delivery: Supabase, Python, GitHub Actions, Vercel
- Agent tooling: Claude Code and Codex skills, plugins, and hooks; MCP; agent evaluation; static analysis; dependency-free Python CLIs
- MBA, Data Analytics concentration — Rowan University
- First-generation Polish-American, based in New Jersey
- 18K+ LinkedIn followers and 3× Top Voice
- Philly sports across all four, golf whenever possible
- 130 memorized digits of Pi, listed on the Pi World Ranking List
