Jeffrey V. Johnson, Ph.D.
Building infrastructure for model evaluation and agent systems
I build software and development infrastructure for reliable agent-assisted work, from native applications and data pipelines to web systems and evaluation tools.
I define tasks, guide and review agent-assisted implementation, investigate failures, and check whether the evidence supports the result.
Seeking research engineering and systems work with meaningful ownership on a specialized team.
Tenured associate professor of mathematics, Kenai Peninsula College/UAA · Homer, Alaska
Selected work
Independent, unpaid projects developed alongside my faculty appointment. The code and case studies show what I built, how I investigated failures, and what the evidence establishes.
- Host Capability Substrate TypeScript prototype for defining what agents may do and recording evidence behind authorization decisions. Public schemas and automated architecture checks make the design inspectable. Examine the public implementation →
- ScopeCam Android microscope application spanning Kotlin, Compose, C++, JNI, and USB camera integration. Built capture and media workflows and diagnosed a camera-replug deadlock across the managed/native boundary. Signed client alpha delivered in June 2026; source is private and recovery retains a resource-leak tradeoff after timeout. Read the native engineering case study →
- Budget Triage Financial evidence workbench with exact monetary handling, provenance, tenant isolation, and human review. A public source snapshot is available; active development is private; the system is not deployed. Review the workbench →
- Agentic-Coding Evaluation Lab Evaluation methodology and a separate verifier that checks whether experiment evidence supports a model-change decision. Public synthetic examples refuse a conclusion when evidence is missing; real-model results remain to be established. Explore the method and evidence →
- Dicee Live multiplayer browser game using Rust/WebAssembly, SvelteKit, Cloudflare Durable Objects, and WebSockets, with public source and a live deployment. Review the delivered system →
Recent upstream contribution: Submitted a Python correction for provider-error diagnostics lost between spawned workers and the parent process, with regression evidence for the proposed change. I responded to maintainer scope feedback with further validation on September 12, 2026. The PR remains under review, awaiting a reply. Review the Inspect Scout pull request →
More relevant work: Governance and agent-control architecture · Agentic Architecture Audit · Full project index