I build local-first AI, automation, and developer tools for practical workflows that need privacy, review, and reliable evidence.
My projects usually sit where scripts are not enough but heavyweight platforms are too much: product discovery, document triage, smart-home audits, cautious file operations, and small-business pre-accounting.
I care about software that is useful before it is flashy: reproducible setup, clear safety boundaries, privacy-aware defaults, and changes a person can inspect before anything important happens.
| Project | What it demonstrates | Verification signal |
|---|---|---|
| TaskSignal | Full-stack AI-assisted product discovery from public developer complaints into evidence-backed software opportunities and Codex-ready MVP prompts. | Public release, CI, release-readiness workflow, Python/FastAPI/Next.js/PostgreSQL |
| MetroPulse Lakehouse | End-to-end urban-mobility data engineering from deterministic source generation through DuckDB bronze/silver/gold layers, FastAPI, and a dependency-free dashboard. | CI, tests, quality gates, architecture documentation, and a local quick start |
| Bokpilot | Local-first pre-accounting assistant for Swedish small businesses, with human-reviewed BAS/VAT suggestions and export drafts. | Sample data, screenshots, unit tests, Python 3.10+ CI |
| Paperless Review Companion | Manifest-first local-model review and triage for private Paperless-ngx archives. | CI, sample fixtures, CLI-first local review flow |
| Home Assistant Audit Toolkit | Read-only Home Assistant audit bundle validation and upload-safe export tooling. | CI, sanitized sample data, privacy-safe export tooling |
| Mac Cleanup Manifest Toolkit | Cautious file cleanup built around inspect, manifest, dry-run, apply, and undo workflows. | CI, dry-run/apply/undo workflow, relative-path reports |
- Local-first defaults for sensitive workflows.
- Human review before external effects.
- Manifest, dry-run, audit trail, and rollback patterns.
- Clear boundaries between experiments, forks, private work, and owned portfolio projects.
- Tools that can be run, inspected, and improved without hidden services.
Making local AI tools more practical for everyday work, and turning messy personal and operational workflows into small, inspectable public projects.