WORK // BUILD LOG
Proof beats positioning.
The portfolio is deliberately mixed: AI infrastructure, developer tooling, private workflow products, live telemetry, and research. I want the systems to show how I think—not just the titles on my résumé.
Featured case studies
Different problems. Same operating style.
Understand the workflow. Model the real constraints. Automate only what earns automation. Make failure visible. Ship something people can actually use.
AI engineering · Context systems
Open sourcePersonal Memory Bank
Persistent project memory governed by authority, freshness, provenance, hooks, diagnostics, semantic drift detection, review workflows, and CI.
Read case studyDeveloper tooling · Local AI
Open sourceAI Code Review Agent
Local multi-agent review with specialist isolation, sanitized diffs, confidence-aware findings, privacy-conscious inference, and multiple developer interfaces.
Read case studyTravel memory product
Private case studySide Quest Atlas
Private small-group travel memory capture designed around 'capture now, enrich later,' group ownership, and clear access boundaries.
Read case studyTelemetry product
Open sourcePitLogic
Client-side BBQ telemetry and analytics with live device integration, alerts, stall detection, forecasting, history, and data portability.
Read case studyPrivate, not hidden
Privacy can be a product decision too.
A private repository is not a weak portfolio artifact when the product contains family history, club data, or personal performance records. The case study exposes the judgment without exposing the people.
Live event workflow
Private case studyTipsy Bunghole
A private whiskey-club product that coordinates blind tasting events across participant, steward, and host roles while preserving institutional memory.
Read case studySports tracking
Private case studyBowling Tracker
A focused bowling product for sessions, scoring, equipment, patterns, leagues, drills, and analytics—designed around the bowler's actual workflow.
Read case studyThe research underneath it
Harness Engineering
Building PMB and ACR kept exposing the same truth: model quality is only one variable. Context, tools, state, authority, evaluation, and execution boundaries decide whether the whole system is trustworthy.