I'm Çınar. I build software by starting with a rough idea, getting it working, and then spending way too much time fixing the parts that annoy me.

Most of what I'm interested in sits somewhere between local-first software, AI experiments, developer tooling, automation, and performance work. I like projects where I can understand the whole path instead of hiding everything behind a hosted API.

Right now most of my public work is in JavaScript/Node.js and Python. I also experiment with local language models, fine-tuning, inference, multimodal tools, and small learning systems on consumer hardware.

A local-first encrypted file vault that uses the user's own Gemini Web session as the remote storage transport.

• Encrypts files locally with AES-256-GCM before provider upload
• Keeps an encrypted local mirror for reliable recovery
• Rebuilds vault records from marked Gemini conversation history
• Uses a local Node.js/Express server with a plain browser frontend
• Includes request guards, provider URL validation, secret scanning, tests, CI and CodeQL
• Licensed under GPL-3.0-only

`JavaScript` `Node.js` `Express` `AES-256-GCM` `Local-first` `Security`

A persistent self-play chess learning engine. This one started as a much smaller experiment and grew into a proper training/evaluation project.

• Parallel CPU self-play
• Alpha-beta + quiescence search
• Transposition tables and move-ordering heuristics
• Temporal-difference evaluator updates
• Paired arena matches with champion gating
• Persistent checkpoints, metrics and legacy checkpoint migration
• Python 3.11+ with tests, Ruff, CI and CodeQL

`Python` `Self-play` `Search` `Game AI` `Training`

Not everything I work on is public. I also spend time testing local models, fine-tuning workflows, quantization, small custom language-model experiments, voice/image/video tooling, and ways to squeeze useful inference out of consumer hardware.

I don't treat those as finished products unless they actually become one.

Mainly: Python, JavaScript, Node.js, HTML/CSS, Git/GitHub
Also around my projects: local AI runtimes, model tooling, test automation, GitHub Actions and Windows development

• Start with something small enough to actually run
• Test the real failure cases instead of only the happy path
• Keep persistent state recoverable
• Prefer local-first designs when they make sense
• Measure regressions instead of assuming a change is better
• Keep iterating after the first working demo

```yaml building:

• cryox-gemini-drive
• cryox-chess-learning

exploring:

• local AI and fine-tuning
• efficient inference on consumer hardware
• model evaluation
• local-first software
• security and reliability

learning_by:

• building
• testing
• breaking things
• fixing them ```