Independent R&D developer working on AI systems, verification, version intelligence, reliability, and knowledge architecture.
I build experimental systems where deterministic computation, measurable evidence, and language-model interpretation are kept deliberately separate.
Exact reconstruction and version intelligence for evolving structured data.
Public experimental release with deterministic reconstruction, SHA-256 verification, temporal/evidence layers, reproducible frozen benchmarks, automated CI, and documented trust boundaries.
Ask the history. Prove the answer.
Measured delayed generalization in a compact PyTorch Transformer.
Public experimental release with executable training code, deterministic smoke tests, automated CI, and a frozen 40,000-step modular-addition run with checksums and functional checkpoint replay.
Deterministic reference simulator for bounded autonomous-control decisions.
Public experimental reference release with runnable TypeScript code, trust-gated decisions, deterministic tests, exact artifact replay, checksums, automated CI, and explicit limitations. Private policies, operational thresholds, production heuristics, and unpublished know-how are not included.
Eight synthetic AI reliability scenarios with reproducible Golden Run 003 artifacts.
Open the live demo · Read the evidence boundary. The public fixtures illustrate safety dispositions; they are not production evidence or a replay of the private implementation.
- BASIS — verified multi-model arbitration, evidence tracking, and structured disagreement;
- AI Clean Layer — validated-memory boundary, provenance, and gated retrieval;
- Human Clarity AI — preserving human agency in AI-assisted reasoning.
LLM never creates metrics; LLM only interprets measured evidence.
I treat reproducibility, provenance, uncertainty, and explicit failure modes as part of the system design rather than as documentation added later.
For project summaries, current status, and publication boundaries:
Open the Independent R&D Portfolio →
Interested in technical and research collaboration around verified AI systems, deterministic evidence layers, version intelligence, and reliability of agentic systems.
Projects shown here are experimental research and evolving prototypes unless explicitly stated otherwise.