At Tessa Labs, I build side projects and experiment with search, speech recognition, and voice technology for Bosnian, Croatian, and Serbian. I built Canto, a meeting recorder and notetaker for iOS, Android, and web, with transcription, summaries, search, and speaker identification.
My speech-recognition research fine-tuned Whisper large-v3 to reach 9.9% WER on FLEURS, compared with 29% for the base model. I also research retrieval for long-context question answering and speaker identification.
At Sudowrite, I built and owned the iOS and Android app and built the agentic chat feature that much of the platform now runs through. I also adapted TipTap for mobile and worked on site performance and the move to TypeScript.
Across earlier product work, I built software including Klu.ai's RLHF Studio, Bloop Gastro, and a voice-to-text app at Clarify. I’m also pursuing a part-time MSc in Computer Science at the University of Liverpool, and I’m drawn to research at the intersection of engineering and the humanities and social sciences.
