Alongside my practice as an attorney in Denizli, Türkiye, I build tools that let AI assistants use Turkish legal sources without fabricating citations, along with Telekumanda, which I wrote to manage AI coding agents from my phone. The source code of all four is open on GitHub under the AGPL-3.0 license.
Case law and legislation research server. Full-text and semantic search with citation checking over 11.1 million court decisions and 305 thousand legislative articles. The firm's day-to-day research infrastructure.
Information server that matches physicians' questions on patient rights, informed consent and health data with official legislation and reasoned high court decisions. It never gives a definitive verdict.
Legislation reader that splits statutes into articles, paragraphs and subparagraphs and detects amended articles; with a PDF library and a study planner, on its own server.
Remote control that manages Claude Code, Codex, OpenCode and Antigravity sessions running on the computer from a phone or a browser: approvals, notifications and a two-sided file manager. The bridge runs on a private network only.
The three tools that work with legal sources share one principle: decisions, dates, docket and decision numbers or article texts are never written from memory. Text is taken from the official source; a source whose full text cannot be found or arrives incomplete is not used, and this is said plainly.