I build software that turns ambiguous input, automation, and complex technical work into something people can inspect, understand, and control.
Daily committed snapshot from GitHub's logged-out public profile view · the signed-in owner's contribution calendar can differ because it may include private activity · exact 365-day window ending on the date printed in the card · no credential or private repository detail requested · a failed refresh keeps the last good snapshot online · Open Studio · Source
My current projects span developer productivity, creative tooling, browser security, puzzle systems, operational evidence, and engineering research. They share a common approach:
- local-first where ownership matters;
- review before consequential automation;
- provenance and evidence instead of unexplained output;
- visible unknown, degraded, stale, and failed states;
- calm interfaces over activity theatre;
- tests, release discipline, accessibility, and recovery as product features.
|
A local-first work operating system that turns notes, transcripts, files, and agent requests into source-linked proposals. People and agents can prepare work, but approval and apply remain explicit capabilities. Now: public beta; stable v0.2.0 and v0.3.0 release candidate; .NET 8, Vue 3, SQLite, SignalR, and review-gated MCP access. |
A local creative workflow studio above ComfyUI for generative images, reference editing, animation, 3D, model research, native-tool handoff, and reproducible experiments. Now: active configured-machine workbench; evolving toward role-aware multi-reference composition, visual workflow authoring, headless parity, qualification, and accepted-asset lineage. |
|
A local-first Chrome MV3 interaction guard for deceptive navigation, risky submissions, DoubleClickjacking, ClickFix-style traps, and other browser actions where visible intent can diverge from the committed result. Now: pre-alpha and controlled dogfooding only; focused on strict authority binding, replayable attack/benign evidence, measured quietness, and release integrity—not efficacy claims. |
An offline-first illustrated puzzle cabinet, mystery collection, and local authoring workshop. It combines versioned logic puzzles, explainable deductions, durable local saves, Wrenmere, Quiet Wing, and a deliberate mobile path. Now: browser/PWA release 0.11.3 with 355 puzzles across thirteen families; Capacitor/Android is planned and architected, not yet claimed as shipped. |
|
A compact operations desk that combines bounded product signals, synthetic checks, release context, and reviewed development evidence into evidence-backed next checks rather than automatic reactions. Now: hosted Cloudflare Worker/D1 Desk; collection is explicitly allowlisted to a small consented Alibi pilot while other integrations remain inert. |
Source-backed GitHub analytics, deterministic README graphics, project-health views, and a live portfolio Studio with explicit freshness, provenance, and no invented rankings. Now: v0.4.0; eight hosted card routes, ten static card types, last-good public snapshots, dark/light bundles, manifests, and an evidence-gated expansion path for maps, scenes, motion, and reviewed projections. |
- Developer Lens — a private, local-first development retrospective with deterministic patterns, evidence walks, hypotheses, a synthetic public showcase, and deliberately redacted export contracts.
- Agent Harness — shared policy, guides, doctoring, adapters, rollout tooling, and evidence contracts for operating coding agents across repositories without turning guidance into blanket command walls.
- N-person Prisoner’s Dilemma simulation — agent-based research on cooperation, reciprocity, network structure, and reinforcement learning; the work behind my SGAI-AI 2025 publication.
- MDviewer, IdleHarbor, RepoScope, and LLM Release Gate — focused utilities and experiments that continue to inform the larger portfolio.
The catalogue below is the last committed credential-free CommitAtlas snapshot. It stays marker-owned and unchanged in this documentation PR; the updated .commitatlas.json will move it to the current six flagships through the existing scheduled or manually dispatched generator after merge.
| Project | Status | Signals/actions |
|---|---|---|
| Taskdeck | Active · CI passing | 1 stars · 305 open issues/PRs; Language: C#; Workflow: ci-required.yml; Release: v0.2.0; Actions: Website · CI · Release · Release download · Docs · Download |
| CommitAtlas | Active · CI stale | 74 open issues/PRs; Language: TypeScript; Workflow: ci.yml; Release: v0.4.0; Actions: Website · CI · Release · Docs |
| IdleHarbor | Active · CI stale | 22 open issues/PRs; Language: C++; Workflow: ci.yml; Release: v0.2.0; Actions: Website · CI · Release · Release download |
| Developer Lens | Active · CI stale | 57 open issues/PRs; Language: TypeScript; Workflow: pages.yml; Actions: Website · CI · Docs |
| LLM Release Gate | Maintenance · CI stale | 2 open issues/PRs; Language: Python; Workflow: ci.yml; Release: v0.1.2; Actions: CI · Release · Download |
| MDviewer | Maintenance · CI stale | 12 open issues/PRs; Language: TypeScript; Workflow: ci.yml; Actions: Website · CI · Docs |
Generated project radar and focused public signals
The project radar and cards are the last committed generated snapshot. Their manifests contain the exact generation time, window, artifact list, sizes, and SHA-256 hashes. A failed refresh leaves that last committed good snapshot available.
GitHub public-profile activity percentages. These are annual profile-view percentages, not exact counts for the Atlas window.
A transparent summary of density, streak, and recent momentum—not a GitHub rank or comparison with other people.
All views use credential-free public GitHub evidence. CommitAtlas requests no private repository names, commits, URLs, or other private details.
- Integrated automated security scanning into roughly 72 AWS CI/CD pipelines using Jenkins, Bash, Groovy, Nmap, and versioned expected-port evidence.
- Published multi-agent simulation and game-theory research through the N-person Prisoner’s Dilemma project and SGAI-AI 2025.
- Build products end to end: product direction, architecture, implementation, tests, CI, release evidence, operations, and user experience.
- Prefer designs where people can see what changed, why it changed, what is uncertain, and how to recover.
- Treat local ownership, least authority, honest degradation, accessibility, and durable evidence as engineering constraints rather than marketing language.
Current toolkit
Languages: C# · TypeScript/JavaScript · Python · C++ · Bash/PowerShell · SQL
Backend & platform: .NET 8 · ASP.NET Core · Node.js · FastAPI · SQLite/Postgres/MySQL · REST · MCP · SignalR · Docker
Frontend: Vue 3 · React · Vite · accessible browser/PWA interfaces
Delivery: GitHub Actions · AWS · Jenkins · Cloudflare Workers/D1 · container workflows · automated security scanning
Applied systems: local-first storage · agent governance · observability · browser security · generative-media workflows · simulation and research tooling
Let’s talk.