Turning AI into working systems.

I design, code, and launch AI systems. I work directly with founders, executive teams, and customers to explore what is possible, rearchitect workflows for speed and scale, and write the software that puts AI into daily production.

Twenty years building software systems. Five years designing and shipping applied AI in production.

Explore production systems ↓

SCALE & EFFICIENCY

Software built for critical work

I build and rearchitect software where operations are critical and speed matters: removing bottlenecks, cutting operational costs, and unlocking capabilities that were previously out of reach.

01 · GLOBAL CAPITAL TRACKING

Cross-border shadow ledger

Built an auditable shadow ledger for an international development bank. Language models parse rules from complex loan agreements, and the system automatically checks disbursements against contract terms across multiple jurisdictions.

02 · DOCUMENT INTELLIGENCE

Contract verification and clause extraction

Built an extraction pipeline that parses enterprise contracts in seconds instead of weeks. The system pinpoints exact clauses down to the character, checks structured data against business rules, and flags uncertain text for human review.

03 · ADVERSARIAL TESTING

Adversarial agents and breaking-point tests

Designed testing suites used by a major advisory firm to stress-test AI agents before release. We use adversarial agents, golden set evaluations, and custom rubrics to find prompt injections, runaway loops, and breaking points before software touches real users.

04 · ENTERPRISE DECISIONS

Safeguards in commercial deal approvals

Connected AI pricing and forecasting models directly into commercial deal approvals. Replaced slow back-and-forth approval chains with automated margin checks, required dual-key sign-offs for exceptions, and logged every calculation to an audit trail.

METHODOLOGY

How I work with teams

I work client-facing from the first idea through launch: exploring what is possible, rearchitecting workflows to run faster with fewer people, building rigorous evaluations, and writing the production code.

01 · EXPLORING WHAT'S POSSIBLE

Idea validation and workflow redesign

For founders and leaders figuring out how to solve a hard problem with AI, or wondering where the breaking points are. I advise on the art of the possible, evaluate whether an idea can actually work in the real world, and rearchitect existing workflows to increase speed, reduce headcount requirements, and unlock capabilities that were previously not available.

02 · EVALUATIONS & CASE STUDIES

Golden sets, rubrics, and adversarial testing

For teams that need to know their models will hold up under pressure. I build evaluation suites directly in your repository: golden test datasets, custom grading rubrics, automated regression tests, and adversarial agents to find prompt injections and tool loops. You get clear case studies and the code fixes needed to pass.

03 · ENGINEERING & ADVISORY

Hands-on coding, client-facing launch, and technical direction

I do not just deliver advice or slide decks. I design the system, write the production code, work directly with your customers and stakeholders, and launch the software. I also provide ongoing technical advisory, helping you make smart vendor choices and defend your architecture in security reviews.

ACTIVE SYSTEMS

Production systems

Four live systems built to move fast, cut server costs, and unlock new capabilities with in-browser compute, local models, and strict execution boundaries.

01 · COLLABORATIVE AGENTS

GroupChatRobot

Ephemeral group chat with private, in-browser transcription.

A zero-dependency group chat platform with temporary room memory. Users join rooms using short codes, and nothing is stored in a permanent database. Speech-to-text runs entirely inside the browser using Whisper WebGPU, so audio never leaves the phone or computer. Strict mention routers ensure the AI only speaks when summoned, eliminating accidental token costs.

Node 24 ESM · Zero npm runtime dependencies Local Whisper WebGPU (no audio egress) Temporary in-memory rooms Exact mention routing

Open groupchatrobot.com → Read build record →

02 · BENCHMARK & SIMULATION

CertArena

Standardized exam platform benchmarking AI reasoning against human professionals.

An exam platform built to benchmark how language models reason through professional certification exams. It runs 228 scenario questions under timed Pearson VUE conditions with a 720-point passing score. Features local Kokoro speech synthesis, automatic citation verification, and zero third-party runtime dependencies.

9 tracks · 228 scenario tests Automated citation verification Local Kokoro audio synthesis (1,132 clips) Zero runtime dependencies

Open aicertarena.com → Read build record →

CertArena Pearson VUE exam simulator showing question stem, countdown timer, and Kokoro speech engine controls.

03 · IN-BROWSER DATA INTELLIGENCE

BlockReport

Fast parcel analytics across 500,000 municipal records, running client-side.

A property research tool that queries over 500,000 city parcels without a central server database. DuckDB-Wasm executes spatial queries, parcel indexing, and ownership history directly inside the user's browser thread. This eliminates database hosting bills, avoids server bottlenecks, and keeps user searches completely private.

500,000+ municipal parcel records Client-side DuckDB-Wasm queries Zero server database bills Zero search tracking

Open blockreport.homes → Read build record →

BlockReport interface: interactive Philadelphia property map, parcel inspector, and DuckDB-Wasm spatial query engine.

04 · AUTONOMOUS SYSTEMS

Penny

Process supervisor and security boundary for autonomous agents.

A control layer that separates AI reasoning from operating system execution across 56 distinct operational domains. The language model only emits structured JSON requests. A local runtime validates the schema, checks cryptographic approval tokens, and executes specific command-line tools with strict exit-code checks.

304 CLI tools · 56 isolated domains Cryptographic approval tokens (model proposes, code enforces) Self-healing process supervisor Strict schema validation gates

Read complete build record →

Penny operating surface: system briefing interface and tool execution telemetry.

Earlier products, algorithms, and retained build records are documented in the archive.

Explore all 15 systems in the project archive →

TRACK RECORD

Engineering background

Making AI reliable is a systems engineering problem, not a prompting trick. It takes disciplined architecture, strict input validation, and twenty years of distributed systems engineering.

APPLIED AI · SINCE 2020

Production ML & agent runtimes

Building machine learning systems and AI runtimes since 2020. Moved enterprise teams from legacy heuristic classifiers to modern tool-calling architectures, stateful agent runtimes, and continuous integration evaluation suites.

GLOBAL SCALE · 27 PRODUCTS IN 11 COUNTRIES

International engineering leadership

Led distributed engineering teams across 11 countries for a global publisher. Shipped 27 international product launches in under 12 months, refactoring multi-terabyte data migration pipelines from two weeks down to four hours.

DISTRIBUTED SYSTEMS · 2000s TO 2010s

High-concurrency infrastructure

Built distributed caching layers, message queues, and low-latency databases for Sony, JackThreads, and high-volume media properties. Built to withstand extreme traffic spikes, strict SLA budgets, and sub-second P99 latency requirements under production load.

FOUNDER & OPERATOR · SINCE 2009

Client-facing product ownership

Over fifteen years building, deploying, and maintaining independent software products from initial customer discovery to production operations. Complete ownership across product design, code, client communication, and systems infrastructure.

CONTACT

Let's talk about your project.

Based in Philadelphia (US Eastern), working with teams worldwide.

Whether you are exploring a new idea, trying to figure out what is possible with AI, or looking to rearchitect an existing workflow to move faster with a smaller team, send me an email. Tell me about your systems, your problems, your project, or what you need to build. I read every note and reply directly with practical next steps.

[email protected] →