I build AI systems you can trust.
Quality gates, self-auditing agents, and composable tooling for AI-native work. I came up in manufacturing, so I run AI like a production line: I turn every mistake into a check the system can't skip, and "done" has to show its receipt.
I run systematic, validation-first trading systems on the same principle.
The throughline: the machine doesn't get the benefit of the doubt. It proves it, or it doesn't ship.
| Repo | What it is |
|---|---|
| andon | Stop fixing the same AI mistake twice — a Lean quality system for AI-assisted work: every defect becomes a permanent countermeasure. |
| verification-discipline | AI verification: a runnable mutation-tested gate, a case study, and a mistakes log with the tests that catch them. |
| primefold-skills-public | Composable, open-source Claude skills: capture, triage, diagnose, gate, write. |
| trading-validation-framework | Strategy validation: out-of-sample splits, Monte Carlo, walk-forward, regime, friction, ablation. |
| pinescript-examples | TradingView PineScript v5 patterns and educational indicators. |
| Project | Contribution |
|---|---|
| BerriAI/litellm | #43196 — fixed doubled reasoning text in Anthropic streaming: independent reproduction on main, a regression test, and cross-provider verification (Anthropic + Bedrock). |
More in private repos: production trading systems, custom GPT configs, and the operating-system internals. Reach me at primefold.tools.
Open to agent-reliability and AI-evals roles, and consulting — collinnerdahl@gmail.com.