Forward Deployed AI Researcher · Agentic AI Engineer

_{Building verifiable, long-horizon agent environments · Enterprise AI & GTM}

Gradia · Snorkel AI · Axiom Consulting

I’m a forward deployed AI researcher and agentic AI engineer building systems that make autonomous behavior observable, replayable, and independently verifiable.

My work sits where agent infrastructure meets experimental science: record what happened, challenge the evaluator, preserve the evidence, and make every important claim reproducible by someone else.

My upstream work is discovered automatically across public repositories outside my account and included only when GitHub also lists me as a contributor. Merged PRs are the accepted-work measure; GitHub-indexed commits are a separate cached attribution signal. Stars and forks describe repository reach, not personal credit.

• MoneyPrinterTurbo: GitHub-listed contributor with 106 GitHub-indexed commits and 105 merged PRs; +8,234 / −738 accepted lines across 245 changed files. Repository reach: 128,021 stars and 20,034 forks.
• openrig: GitHub-listed contributor with 82 GitHub-indexed commits and 86 merged PRs; +6,893 / −554 accepted lines across 248 changed files. Repository reach: 3,936 stars and 261 forks.
• agency-agents: GitHub-listed contributor with 65 GitHub-indexed commits and 57 merged PRs; +1,894 / −528 accepted lines across 92 changed files. Repository reach: 155,717 stars and 25,147 forks.
• Ruflo: GitHub-listed contributor with 89 GitHub-indexed commits and 51 merged PRs; +6,041 / −801 accepted lines across 195 changed files. Repository reach: 73,694 stars and 8,755 forks.
• Agentic-QE: GitHub-listed contributor with 93 GitHub-indexed commits and 50 merged PRs; +12,870 / −2,288 accepted lines across 313 changed files. Repository reach: 489 stars and 96 forks.
• CowAgent: GitHub-listed contributor with 29 GitHub-indexed commits and 27 merged PRs; +2,829 / −318 accepted lines across 60 changed files. Repository reach: 47,213 stars and 10,377 forks.
• VoiceStudio: GitHub-listed contributor with 38 GitHub-indexed commits and 18 merged PRs; +2,522 / −134 accepted lines across 78 changed files. Repository reach: 51,596 stars and 5,753 forks.
• iFixAi: GitHub-listed contributor with 11 GitHub-indexed commits and 11 merged PRs; +849 / −83 accepted lines across 26 changed files. Repository reach: 18,812 stars and 1,369 forks.
• benchmark-radar: GitHub-listed contributor with 1 GitHub-indexed commit and 1 merged PR; +85 / −7 accepted lines across 4 changed files. Repository reach: 270 stars and 45 forks.
• imajev: GitHub-listed contributor with 1 GitHub-indexed commit and 1 merged PR; +164 / −2 accepted lines across 4 changed files. Repository reach: 205 stars and 22 forks.

Owned public source is a separate signal: GitHub lists me as the repository owner. These projects are discovered automatically from my public, non-fork, non-archived repositories, then ranked by stars, recent activity, and GitHub-attributed commits. Ownership is not counted as upstream contributor credit.

• ProofSeal: public owner repository with 13 GitHub-indexed commits. Repository reach: 2 stars and 0 forks; last pushed 2026-06-13.
• TestLore: public owner repository with 99 GitHub-indexed commits. Repository reach: 1 star and 0 forks; last pushed 2026-10-01.
• ghostpart: public owner repository with 13 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-10-01.
• nerve: public owner repository with 3 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-09-28.
• Gradia Guard: public owner repository with 10 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-09-25.
• gradia-wind-tunnel: public owner repository with 83 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-09-03.
• gradia-universes-work-sample: public owner repository with 34 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-09-03.
• gradia-reward-loop: public owner repository with 33 GitHub-indexed commits. Repository reach: 0 stars and 0 forks; last pushed 2026-09-03.

_{Showing 8 of 19 qualifying owned public repositories · browse every public source repository · complete evidence retained in machine-readable data}

_{Last verified 2026-10-02 UTC · visual + evidence refreshed every 30 minutes by GitHub Actions · machine-readable evidence}

Agentic Power asks a practical question: how much accepted skilled work can one hour of human direction command?

I design for that multiplier, but I won’t publish one without exposing its scope, evidence, assumptions, and uncertainty. My operating model is to delegate meaningful responsibility, compose the right tools and context, verify the result, then use the evidence to improve the next run.

AP ≈ 76.9× (operator-calibrated provisional estimate): approximately 10,981.8 skilled Human-Equivalent Hours, or 274.5 engineer-weeks, divided by 142.9 operator-estimated human-direction hours. The transparent scenario range is 48.3× to 123.8×.

The evidence base since January 2025 combines 407 merged upstream PRs with 6,532 authored, non-merge default-branch commits across 109 currently accessible repositories: personal, private, employer, and open-source alike. Work such as Gradia is included only as a redacted aggregate: no repository names, commit messages, code, links, employer, or client details are published.

The conservative GitHub activity proxy produces 1,650.6 direction hours because it assigns attention to individual commits and repository-months. Operator recall is 1.5–2.0 active direction hours per week; across 81.7 weeks, that calibrates the denominator to 122.5–163.3 hours, with 142.9 as the midpoint.

Direction includes active briefing, steering, reviewing, correcting, and coordinating. It excludes agent runtime and waiting. The calibration is operator-estimated rather than reconstructed from time logs, so this remains a transparent scenario, not a completed Full Evidence Audit.

_{Framework and formula · calculation evidence · public upstream evidence and visuals refreshed every 30 minutes; redacted repository snapshot retained until a private read credential is available}

• Direct: goals, constraints, acceptance criteria, and escalation rules stay explicit.
• Orchestrate: models, specialist agents, reusable skills, tools, memory, and loops become one working system.
• Verify: Gradia Guard, ProofSeal, and replayable evaluations separate completed work from plausible-looking activity.
• Improve: Wind Tunnel and Reward Loop turn measured failure into a better next run.

_{Framework concept by Dr. Mark Allen / HeroForge.AI. Visual and operating-model adaptation are original to this profile.}

Can we trust the record?      → proof-bound execution evidence
Can we trust the score?       → adversarial evaluator stress testing
Can we reproduce the result?  → deterministic replay + committed artifacts
Can an agent remember?        → regression memory before commit

The thread through all of it is simple: an AI system should be able to show its work without asking you to trust the AI system.

• Gradia Universes — proof-carrying, interruption-capable synthetic agent worlds with deterministic replay.
• Gradia Reward Loop — oracle-witnessed reward-hacking experiments and a replay-verified paired-GRPO diagnostic.

I build communities as deliberately as systems. I lead AI Tinkerers Charlotte, a demo-first room for people shipping real AI systems, and serve as Regional Ambassador for the Agentics Foundation, connecting builders working on practical agentic AI.