Worked on stabilizing and enhancing Python-based audio processing and speech recognition systems across multiple repositories, including davideuler/index-tts, QuentinFuxa/WhisperLiveKit, and Blaizzy/mlx-audio. Improved build reliability and maintainability in index-tts by refactoring imports, updating ignore files, and resolving packaging issues to streamline dependency management. In WhisperLiveKit, addressed thread safety by implementing a thread-safe singleton and model locking to prevent race conditions during concurrent access. Developed a demonstration script for Qwen3-ASR in mlx-audio, showcasing a two-pass transcription workflow with forced alignment for word-level accuracy. Applied skills in Python scripting, concurrent programming, and package management throughout the work.