Full-stack product engineer. I build early-stage, AI-native products end to end : web, mobile, backend and the product layer around the model : and I state how every figure I quote was measured.
- Founder, LILA (Feb 2026 to present) : a proactive personal AI whose hardest problem is knowing when not to speak: one delivery gate with twelve ordered checks, every rejection logged with its reason, and an evaluation harness that scores silence as its own dimension. Sole author, ~9,500 commits, sixteen languages, store-ready with the release deliberately held.
- Founding engineer, Cendra (Aug 2025 to Jul 2026) : primary author of the web application, sole author of both mobile apps, owner of the notification and real-time system. $1M seed led by Revo Capital, February 2026.
- Founding frontend engineer, ReadyFly (Jun 2025 to Jun 2026) : Maya, an AI career agent.
All five install from npm, each published by its own release workflow with build provenance; proactive-gate's Python sibling is on PyPI the same way. On the evening they were published an audit found seven defects, six of them reproduced by running the tools: two failed outright on a decomposed filename, the linter's score was meaningless outside the Latin alphabet, and a name written in Turkish capitals could not match itself. All seven are fixed, each with a test built from the reproduction.
| Package | What it does | Install |
|---|---|---|
| ai-slop-linter | A linter for the tells of machine writing in commit messages, PR descriptions and READMEs: twenty sourced rules, a score per file, safe fixes, an Action, a commit hook and an agent skill. It lists tells; it never guesses who wrote the text. Precision measured on a fixed corpus: pre-model public-domain prose grades A or B, unedited model output grades F; the score is computed in any script, not only the Latin alphabet.  |
npx ai-slop-linter README.md |
| product-engineer | A pack of four agent skills: your coding agent restates work as a customer outcome, writes a plain-language "For the customer" block in every commit and PR, never calls unobserved work done, and writes release notes people can read; check ships as a commitlint plugin and a pre-commit hook. Eight tasks run bare and with the skill, every transcript in the repository, and each skill states what was not measured about it.  |
npx skills add Bubblegunn/product-engineer |
| proactive-gate | LILA's delivery gate as a zero-dependency library: twelve ordered checks, a reason on every rejection, a replay CLI; policies as JSON, fourteen presets with sources, adapters for the AI SDK, Mastra, LangChain and OpenAI Agents, a Python sibling on the same fixtures, and quiet hours that differ per weekday for a working week that is not Monday to Friday.  on npm, and on PyPI at the same version, with a docs site |
npm i proactive-gate |
| workproof | A verifiable engineering report from a private git repository, without showing code: thirteen git-derived figures with the command behind each, bots and generated files excluded, a canonical-JSON hash, check, verify and a signed attest; the README publishes how each figure responds to gaming; calendar figures use the author's own weeks.  |
npx workproof |
| surviving-lines | Share of surviving lines next to commit share over a deterministic file sample; the method behind the ownership essay, with a Markdown table an outside contributor added.  |
npx surviving-lines |
Every release of all five is archived on Zenodo with a permanent identifier, so a paper or a report can cite the exact code it ran: proactive-gate, product-engineer, workproof, surviving-lines, ai-slop-linter.
| Project | Change | Outcome |
|---|---|---|
| langchain-ai/openwiki #635 | Bundled skills could not sync from read-only installations | Merged, released in v0.3.3 with a named changelog credit |
| langchain-ai/openwiki #703 | Onboarding tests wrote into the developer's real config directory on Windows | Merged |
| mastra-ai/mastra #22903 | runEvals gate loops broke the trajectory-scorer contract and hid the failure as a silent 0 |
Approved and merged; shipped in @mastra/core@1.64.0 |
Each carried red→green regression tests; on the mastra one, the reviewer reproduced the regression check himself before approving.
Eleven essays on building this way : efe-genc-portfolio.vercel.app/writing : including The Spec Is the Fast Path, Knowing When Not to Speak and Showing Ownership When the Repositories Are Private. The /stats page prints every number with its scope and method.
Five merged pull requests from three people I had never spoken to arrived on these repositories in their first day. Most of my product work lives in private repositories (LILA, and my employers'); the five packages above are the parts of that work I could make public and reusable. Ankara, Türkiye · open to UK relocation · efegenc95@gmail.com · LinkedIn