Toby
Iverson

I build systems that check their own work. By day I'm an automation engineer in Minneapolis — Python, SQL, and the unglamorous plumbing that keeps platforms honest. The rest of the time I ship AI systems under the BytesFromToby flag: an autonomous operator triaging GitHub issues live in production, a strategy game where 28 AI factions negotiate under a deterministic rules engine, and pipelines where the spec — not the vibes — is the source of truth.

The thread through all of it: verification beats trust. Software should be able to prove it did what it claims, whether it's a Tkinter app parsing a gigabyte of logs or an LLM whose every promise is checked by 638 tests before it becomes state.

1. RepoRouterAn autonomous AI operator triaging GitHub issues live in production on a local 8B model — categorizes, checks claims against the repo's own docs, then replies, labels, closes, or escalates. Safety enforced by the harness, guardrails pinned by pytest. AI OperatorIn Production
2. Polis28 AI-driven factions contest a Greek city-state through live LLM negotiation; agreed terms parse into structured deals a deterministic engine enforces. A living public, elections, conspiracies — a reign you can win or lose. The model proposes; the engine disposes. AI + Game Engine638 tests · v0.3.2
3. plumblineIdea to verified, working code through eight single-responsibility AI-agent skills and two unattended orchestrators. The spec is the single source of truth; the contract that binds the skills is itself audited in CI. Built Polis end-to-end. AI Agentsv1.0
4. Log AnalyzerDesktop log analysis with multi-file timestamp merging, known-issue matching, color-coded preview, and a reusable report builder. Pure Python 3 + Tkinter — zero external dependencies. Python★ 3 · v1.0
5. DrugResearcherA folder-based AI research analyst that maps drug-development pipelines, tiers every source, then audits its own findings with a hostile reviewer before reporting back. AI ResearchMIT