01
AI evaluation and benchmark design
Controlled model and agent evaluations using realistic tasks, explicit acceptance criteria, machine or judgment-based oracles, isolated trials, and documented limitations.
Evidence before assurance
I investigate whether AI-assisted systems actually do the work they claim to do.
I design tests, inspect connected workflows, direct and verify agent-assisted implementation, and turn technical evidence into decisions about what can be trusted, what needs correction, and what must remain under human review.
Capabilities
Quality is an evidence problem
My work begins with the acceptance boundary: what would need to be true before the result could be used, delivered, or trusted?
02
Systems QA and acceptance
Testing whether components produce a usable connected workflow, investigating false completion and integration failures, and defining the evidence required before work is accepted.
03
Technical investigation and research
Inspecting source implementations, comparing tools and architectures, tracing provenance, testing claims, and translating findings into practical decisions.
04
Human-supervised workflow design
Designing AI-assisted workflows, approval boundaries, recovery behavior, and knowledge systems that preserve human authority and durable records.
Selected work
Inspect the work, not the adjective
Public evaluation portfolio
AI Evaluation and Benchmark Engineering
Five inspectable studies covering benchmark design, controlled model comparisons, professional-report review, agent reliability, verification methods, and bounded claim-making.
Public repository · Runnable or inspectable evidence
Private application · Public case description
Studio
An agent-assisted desktop media-production application used to investigate product direction, acceptance design, agent supervision, integration testing, and the difference between passing components and a usable connected workflow.
Under development · Source remains private
Operational personal system
Cross-Agent Continuity and Knowledge System
A working system combining searchable conversation archives, local semantic retrieval, portable Markdown records, mobile capture, synchronization, scheduled reporting, and explicit information-governance rules.
In daily use · Multi-user product work is separate
Selected writing
Technical evidence, explained plainly
The AI Finished. The Work Wasn’t Done.
Why completed output, passing checks, cost, elapsed time, and remaining human work belong in the same evaluation.
Medium · AI evaluation and workflow reliability
Your Autonomous Agent Still Needs a Manager
Why successful completion does not prove that an autonomous agent worked independently, and why scope drift, supervision, and unnecessary implementation belong in the result.
Medium · Agent reliability and supervision
The Administrative Work Between Doing the Work
How conversation search, durable notes, synchronization, mobile capture, and scheduled reporting became one continuity system.
Medium · Knowledge systems and workflow design
Ways to work with me
Roles and bounded engagements
Employment and contract roles
I am open to remote full-time and contract work in AI quality, agent evaluation, systems QA, technical research, and technical writing.
Bounded evaluation engagements
For defined consulting projects, I evaluate sanitized AI-assisted workflows and provide written findings covering tested capabilities, failure patterns, costs, limitations, and remaining human-review work.
Contact
Start with the role, problem, or question
For employment, contract work, technical research, writing, or a bounded evaluation project, email a brief, non-confidential description of what you need.