Projects
The through-line is legible, externally-grounded systems. The center of gravity is AI-safety verification; the rest is the technical range it grows from — labelled honestly by maturity. The open-source tools are linked; the research code and evidence are available to reviewers on request.
AI safety & verification
When can AI check its own work? A verifier's value is gated by its own competence (corr −0.87, replicated ×3, matching a law derived from first principles), "independent" models share error basins (same wrong answer ρ ≈ 0.55 when both miss), and self-repair crosses failure boundaries only with a foothold. Local, replicated; frontier validation is the proposed next step.
A multi-project program showing simple mechanisms at system boundaries match or beat far larger internal-model approaches — code repair (the "multi-file wall"), continual learning, and model routing. Reported with five honest negatives and three documented overclaim retractions.
AI systems & tools
Local OpenAI-compatible execution gateway (~13,300 LOC, 133 tests): tiered model routing, async task management, session/provenance records, and a containment validator (100% adversarial detection / 0% false-positive across 200 trials).
A git-native, test-gated wrapper for LLM code-repair loops, packaged on PyPI — restores reliable multi-file repair where a bare loop silently reverts good fixes. Benchmarked on 22 purpose-built workspaces.
A keyword router that classifies each request at the boundary and routes to the cheapest adequate model tier — 23–65% measured cost reduction at 95–96% quality retention. Open source (MIT).
Research substrate + the boundary-mechanism controls used across the program (the "internal-model alternative" baseline). Packaged Python.
Mathematics
A computational study of Robin's inequality for colossally-abundant numbers — mapping a gap-decay rate and its connection to the Riemann critical line across 38 numbered experiments. Honest about being proof-adjacent, not a proof.
Climate & energy
A multi-domain optimizer spanning energy infrastructure, industrial transition, critical materials, and ecology — multi-period planning with a discovery loop and a 30+ command CLI.
Artificial life & emergence
Auditable open-ended evolution paired with mechanistic interpretability — measuring emergence rather than asserting it, via a "pairing principle" for observer-relative novelty.
Emergence and complexity in minimal substrates — testing whether structure arises through compression and dynamics rather than top-down design.
An agent-based study of how different mind designs survive resource crises — a "parliament of subminds" proves more resilient than monolithic designs.
The open tools are on GitHub. The papers, raw results, and reproduction bundles are available to reviewers and collaborators on request.
Get in touch