Software engineer at vivenu, building product systems, developer tools, and AI-heavy delivery workflows across work and private projects.
Website · GitHub · LinkedIn · X
I work on production software at vivenu and keep a private/public builder loop around tools, media products, and agent-heavy development. The interesting bit is not that I use AI; it is that I track where it changes the shape of the work.
| What I want to show | Evidence |
|---|---|
| still actively building outside work | 43 repos pushed in 2026, including public tools and private product systems |
| AI is part of the operating model | 952 tracked Codex threads and 17.47B tracked Codex tokens |
| current preference is opinionated | Codex App for stable long-running work, Pi for raw/custom harness work, Cursor Pro for editor-side AI |
| Project | What it is | Signal |
|---|---|---|
| pi-tools | custom Pi package for context, prompts, notifications, stash, and compaction | public, 9 tracked Pi sessions |
| qwer-q | typed, docker-first message queue with schema registry and dashboard | public, pushed 2026-05-21 |
| passepartout | photography export tool for framed/social-ready image output | public, 10 recent Codex threads |
zero / zero-sense |
language and agent-development exploration | top recent Codex focus |
| sandcastle | TypeScript orchestration for sandboxed coding agents | public agent tooling |
| aurora | Convex-first realtime community platform | 1.96B tracked Codex tokens |
| VeraMint | mobile news product with curated journalism and offline reading | private product work |
| Picalyze | photography analysis workflow and product exploration | private product work |
| Layer | Tools |
|---|---|
| Product | TypeScript, Bun, React, Next.js, Astro, Expo, Convex, PostgreSQL |
| Systems | Go, Rust, Docker |
| Delivery | Git, Graphite, worktrees, PR review loops, Playwright, Biome, Vercel, Netlify |
| AI | Codex App, GPT-5.5, local agent skills, Pi + pi-tools, GPT Image (gpt-image-2), Cursor Pro |
Snapshot as of 2026-05-25. The full evidence trail lives in docs/profile-fact-audit-2026-05-24.md.
| Codex 952 threads 17.47B tokens |
GitHub 2,399 contributions 160 PRs |
Pi 35 sessions 10,934 messages |
Cursor 71,463 ACP sessions 346 code hashes |
| Month | GitHub | Codex | Tokens | Claude | Pi | Cursor |
|---|---|---|---|---|---|---|
| 2026-05 | 354 | 118 | 4008.9M | 48 | 3 | 70208 |
| 2026-04 | 167 | 481 | 7521.4M | 46 | 32 | 1255 |
| 2026-03 | 80 | 178 | 4171.5M | 55 | - | - |
| Tool | Why it is here |
|---|---|
Codex App |
default lane for stable, long-running coding work on GPT-5.5 |
Pi |
cleaner/rawer harness and custom pi-tools behavior |
GPT Image |
UI direction, visual assets, and interface exploration |
Cursor Pro |
editor-side AI and current Claude access path |
Claude / Opencode Go |
historical telemetry only, not active standalone defaults |
More evidence
- Codex openers show planning, research, agent, frontend, docs, PR, test, bugfix, review, and worktree patterns.
- The workflow shifted from Claude-heavy exploration in January 2026 to Codex-heavy implementation from February onward.
- Cursor ACP metadata is useful as footprint evidence, but too noisy for project priority.
- Refresh scripts: dashboard, fact audit, claim check.
