A swipe-based true/false trivia experiment; the real subject is not the game but making a client-computed score verifiable through a signature.
Laravel PHP PostgreSQL
+6 more
Labs
The open notebook of unfinished work
Not every idea becomes a product. What I try goes here as it happened — what I tried, what came out of it, what happened next.
- 9
- experiments
- 4
- active
- 5
- graduated
- 41
- technologies
What I am after
The experiments are not scattered; they cluster on four axes. An empty axis means no published entry yet — the direction has not changed.
-
AI-assisted development
Where the LLM enters the development loop: code review, static analysis mixed with models, deterministic filters instead of prompts.
2 experiments
-
Distributed systems & messaging
Queues, outbox, idempotent consumption and language-agnostic message contracts.
1 experiments
-
Developer tooling & CI
CLIs, architecture boundaries enforced at build time, automation that measures code health.
1 experiments
-
Automation & data pipelines
Scheduled jobs, integration pipelines and operational data flows.
3 experiments
On the bench
A Claude Code skill that analyses a Sentry issue against the working tree, proposes the fix as a diff plus a regression test, and touches no file until you approve it.
- Started
- July 2026 — Ongoing
- What I am after
- AI-assisted development
- Technologies
- Python · Bash · Sentry API · sentry-cli · MCP · Claude Code
Ongoing experiments
What is still in the lab: on the bench, shelved, or abandoned. The ones that became products have their own section below.
Active Experiments
4Open-ended experiments I am still working on.
A Reddit-like community experiment where every vehicle make and model has its own club, holding ICE and EV under one roof; API, web and mobile move in parallel.
Laravel PHP PostgreSQL
+6 more
A Claude Code skill that analyses a Sentry issue against the working tree, proposes the fix as a diff plus a regression test, and touches no file until you approve it.
Python Bash Sentry API
+3 more
A content pipeline experiment that screens software and AI news with a local Ollama model, hands the survivors up to DeepSeek, Gemini, OpenAI and Claude, and drafts and publishes Turkish and English syntheses end to end.
TypeScript Node.js Prisma
+8 more
The life of an experiment
Every entry sits in one of four states. I do not hide the state: an abandoned experiment carries as much information as a running one.
- active Active
- Open-ended experiments I am still working on.
- graduated Graduated
- Experiments that outgrew the lab and became products.
- shelved Shelved
- Paused for now, but not closed for good.
- abandoned Abandoned
- Tried, did not work out. They stay here for the lesson.
From experiment to product
Some experiments leave the lab and move into the portfolio. Which one ended up where stays on the record here.
- Two languages sharing one queue without the serialization lock-in BabelQueue
- Screen lock by Bluetooth proximity: present/absent is not enough SignalLock
- Local code review without letting the diff leave the machine CommitBrief
- Personal and business finance on one data model Parantaj
- PHP packages that stand alone and still work together InitPHP
If you want to follow along
New entries arrive over RSS and most of the code is open on GitHub. If there is something worth trying together, just write.