Lab
Active experiments in AI systems, infrastructure, and tooling.
Orka
WIPTreats a pipeline as a composite skill — the DAG is the skill, written in SKILL.md (a cross-vendor open format). Author skills in any AI chat, and Orka becomes the management, execution, and observability layer: visual canvas, scheduling, Mac-native outputs (Apple Notes, iCloud, webhooks) that cloud tools cannot reach.
JobAgent
WIPAI agent that finds jobs, tailors your resume, and prepares a complete application package — in 3 minutes per app. Uses Claude Code + MCP servers for job scraping, JD analysis, and expert-guided resume rewriting.
Harness Design Experiment
CompletedTested 3 multi-agent harness architectures (Solo, Generator+Evaluator, Planner+Generator+Evaluator) on the same task to measure quality, time, and reliability differences. Found that a 3-agent harness with browser-based evaluation catches bugs that code review misses.