I build clean-room public artifacts for reproducible AI evaluation, auditability, and human-reviewed system design.

I build reproducible AI governance benchmarks for agentic systems, with a focus on authority boundaries, tool-use safety, human sovereignty, and whether AI assistance strengthens or weakens human judgment.

Most of my production research is private. This GitHub contains clean-room, educational, and redacted artifacts that show my methods without exposing proprietary systems, private schemas, private logs, private records, or production architecture.

My goal with these repos is simple: make AI governance claims inspectable instead of just aspirational.

The clearest path through my public work is:

1. Public AI Governance Portfolio — the short map connecting the public research arc.
2. Case Study: AI Governance Benchmarks — how I turned an abstract governance idea into a bounded, runnable, tested artifact.
3. Evaluation Stack Overview — the two-layer stack: agent boundary safety and human sovereignty.
4. AI Governance Benchmarks — synthetic inputs, governance scorer, reproducible report, and tests.
5. Agent Action Audit Template — schema-backed synthetic action records, blocked-action examples, and human-review metadata.

Method line:

synthetic inputs -> governance scorer -> reproducible report -> tests pass

I recently expanded ai-governance-benchmarks into a two-layer public evaluation stack: agent boundary safety and human sovereignty.

Status:

• CI green
• 21 tests passing
• deterministic rule-based scorer
• synthetic boundary cases
• generated sample report
• explicit claims and limitations
• anti-overclaim language guard

Repo: https://github.com/tombudd/ai-governance-benchmarks

For the full public evidence map, start with Public AI Governance Portfolio.

My work spans computational cognition, games, humanistic computing, local-first tooling, and public-safe research artifacts.

The common thread is simple: I turn abstract ideas into small, reviewable software artifacts that people can inspect, run, test, or critique.

Some of my public work focuses on AI governance, but the broader pattern is systems design: building tools and demos that make complex ideas more understandable without exposing private architecture or overstating what the work shows.

• Computational cognition — active inference, prediction, uncertainty, homeostasis, agency, and adaptive behavior.
• Games and strategic environments — play, multi-agent decision spaces, learning under constraints, and game-like evaluation settings.
• Humanistic computing — interactive software around literature, culture, education, interpretation, and meaning-making.
• Local-first tools — lightweight systems for research, review, evidence tracking, and human-in-the-loop workflows.
• Public-safe artifacts — clean-room demos, primers, benchmarks, and case studies that show method without exposing private systems.
• Product and portfolio infrastructure — websites, project indexes, prototypes, and deployed surfaces that make research easier to navigate.

I like building systems that are useful, understandable, and responsibly bounded — tools that help people think better, learn faster, repair mistakes, and make decisions with more clarity.

• Active Inference Primer — a minimal educational primer for thinking about prediction, agency, uncertainty, and adaptive systems.
• OpenSpiel — game and multi-agent environment work; useful for studying strategy, learning, coordination, and bounded decision-making.
• Quantum AI Experiments — supporting exploratory sandbox for simulator-first quantum/AI-adjacent experiments with explicit claim boundaries.

AI Governance Benchmarks — clean-room public benchmark portfolio for AI governance evaluation.

Shows:

• synthetic benchmark design
• reproducible local runners
• rule-based scoring
• expected reports
• pytest + GitHub Actions verification
• public/private disclosure discipline

Start with:

• Evaluation Stack Overview
• Project Status
• Recruiter Summary
• Hiring Proof Map
• Case Study

Agent Action Audit Template — clean-room templates for synthetic agent action receipts, blocked-action examples, human-review metadata, JSON schemas, and pytest validation.

Shows:

• agent action audit trail design
• safe vs. blocked action records
• human review checkpoint modeling
• schema-backed receipt validation
• synthetic-only public boundary discipline

• AI Governance Benchmarks — clean-room benchmark portfolio with synthetic Constitutional Adherence and Accountability Completeness evaluations, local runners, scoring, expected reports, tests, and CI.
• Agent Action Audit Template — clean-room operational governance template with synthetic safe and blocked action receipts, human-review metadata, schemas, tests, and CI.
• Human-AI Governance Lab — clean-room workflow for lightweight risk classification, human approval gates, and reproducible audit receipts.
• Quantum AI Experiments — exploratory quantum-circuit experiments related to optimization, inference, and noisy-hardware testing.

These repositories are intentionally bounded. They are public proof-of-method artifacts, not disclosures of private systems.

My work focuses on a practical question:

How do we make AI systems that can act usefully while remaining inspectable, bounded, corrigible, and accountable at runtime?

I am especially interested in:

• Governance mechanisms that execute, not just policies that describe.
• Decision receipts that can be independently audited.
• Evaluation suites for adversarial pressure and boundary stability.
• Human-centered AI systems that preserve agency, dignity, and institutional accountability.
• Safe public documentation strategies for private AI safety research.

I am the founder of ResoVerse, where my private production research informs the public-safe methods and proof artifacts shared here.

I am open to serious collaborators in AI safety, governance evaluation, red-teaming, systems architecture, and applied research.

→ tombudd.com/get-involved — I respond personally.