Security research · Risk and compliance · AI systems
Combat Marine veteran and independent builder. I develop tools that make experiments easier to reproduce, inspect, and challenge.
Before moving into tech I worked in EHS and compliance, running inspections, incident investigations, corrective-action tracking and training. I also do authorized bug bounty research on Bugcrowd.
My work spans agent memory, finite mathematics, and security testing. I care about the gap between an interesting result and evidence someone else can check.
rapidbm25: a Rust drop-in replacement for the abandoned rank_bm25 library, with bit-for-bit identical scores and 1,000x+ faster search. You can install it with pip install rapidbm25. It's also listed on the official MCP Registry so AI agents can use it directly.
| Project | Focus |
|---|---|
| Finite Congruence Checks | Validation of OpenAI math's finite congruence graph criterion through bounded experiments and Lean proof checks |
| Oscillink Safety Ops | Independent safety supervision for AI-controlled industrial equipment |
| Causal Memory Routing Lab | Synthetic experiments in routing, confidence, memory, and forgetting |
| PrepQC Exploitation | Reproducible tests for post-quantum cryptography |
| BountyBreak | Local-first scope, asset and evidence tracking for authorized bug bounty work |
| Oscillink Agent | Cited project memory and continuity for coding agents |
- Keep runnable experiments, source versions, and evidence together.
- Use independent checks and document the limits of each result.
- Retain negative results and corrections as part of the research record.
- Use AI-assisted development with explicit verification and review.
These repositories contain research tools and experiments. Each project's documentation describes its maturity, tested scope, and remaining work.
For a question, reproduction result, or technical discussion, open an issue in the relevant repository.
Earlier simulation work in this repository: SimLab setup and analysis.



