Independent developer building practical AI agent workflows and productivity tools.
I turn repeatable knowledge work into workflows that are explicit, testable, recoverable, and useful outside a demo.
- Spec-driven AI development workflows for Claude Code and Codex
- Evidence-first research and knowledge workflows
- Local-first Chrome extensions and developer utilities
- Small AI products focused on concrete user problems
| Project | What it does |
|---|---|
| BYZ | Business-first coding agent built on Pi, with isolated workflows and traceable runtime foundations |
| CM Workflow | Spec-driven AI development workflow for Codex and Claude Code |
| Zero Workflow | Evidence-first research workflow that collects, cross-checks, cites, and verifies sources |
| LinkAtlas | Local-first private bookmark workspace built with Next.js, SQLite, and Drizzle |
| JSON Keeper | Privacy-first JSON workbench with big-int safety, JSONPath, schema inference, and validation |
| DiffLens | Private in-browser text comparison with JSON normalization, move detection, and patch export |
- Workflow over prompt — durable state, explicit gates, and resumable execution
- Verification over vibes — tests, review loops, and observable outcomes
- Local-first when possible — less data exposure and fewer hidden dependencies
- Products over demos — packaging, failure handling, and documentation matter
Building Codex-native AI Agent workflows and practical productivity tools, while keeping Claude Code compatibility through one shared source of truth.




