I build private, sovereign AI: voice and document systems that keep sensitive data where it belongs (in the browser, in-country, or on your own servers) while still being useful.
Lately that means Turkish legal AI, APAC speech infrastructure and supervised voice agents. Before that, a lot of WebRTC and telephony.
- Katip Yerel: Turkish KVKK redaction that runs entirely in your browser. Finds and masks T.C. kimlik no, IBAN, names, addresses and health data in PDFs, Word, UDF and scans. 757 of 757 values found on an 80-document blind set (with the BERTurk tagger), 0 network requests after a file is opened. Live demo · 🤗 Space · 🤗 BERTurk model · 🤗 tiny model
- Makimoto Kawa (contributor): open-source, Singapore-hosted transcription for conversational AI under APAC data-residency rules. I built the transcription playground, reference client and quickstart for the July 2026 launch. makimoto.ai
- Farend (consulting project, private, in progress): an AI that makes your phone calls for you, supervised: goal → practice call → real call → answer, with consent-gated recording and quote-grounded reports.
- Lu for Delta: an experimental in-game AI assistant in the pause menu of the Delta iOS emulator (Swift).
- WebRTC & telephony: webrtc-ring-to-forwarding, voxbone-webrtc-drachtio, voxrtcapi.
Python · TypeScript · Swift · ONNX / small on-device models · Pyodide & WebAssembly · jambonz / SIP / WebRTC · Docker
Open an issue on any repo, or visit fikrifirat.com. Interested in roles and collaborations around privacy-first AI, voice AI and legal/regulated-sector AI.




