Back to Projects
AI & ML, Desktop AppCompleted
MeeTrack: Local-First Meeting Intelligence & Management
A 100% local-first meeting intelligence platform featuring automated smart note-taking, discussion summaries, and action item tracking, powered by on-device speech separation and speaker diarization.
Key Details
High-Throughput Non-blocking Audio Pipeline: Implemented in Rust using Windows WASAPI loopback, supporting simultaneous capture from microphone, system sound, and specific running applications with zero packet drop.
On-Device Speech Intelligence Workflow: Combines Silero VAD (ONNX) for voice segmentation, int8 Overlap Detector to spot concurrent speakers, and SepFormer for neural audio source separation into discrete stems.
Persistent Speaker Identity Engine: Employs voice embeddings and centroid clustering to recognize participants across sessions, resolve unknown speakers, and retroactively link speech fragments without cloud dependencies.
Multi-Threaded Producer-Consumer Architecture: Decouples continuous audio capture from AI worker threads and inference tasks, guaranteeing 60+ FPS UI responsiveness and continuous recording integrity.
Modern Glassmorphic Desktop Interface: Built with Next.js 15, React 19, Tauri v2 event-driven IPC, and an audio-reactive Lava Lamp visualizer powered by Three.js GLSL raymarching shaders.
Highlights
100% Local-First & Zero Cloud Dependency
Non-blocking Windows WASAPI Audio Pipeline in Rust
On-device INT8 Overlap Detection & SepFormer Speech Separation
Persistent Speaker Identity via Vector Centroid Clustering
Audio-reactive Lava Lamp Visualizer with Three.js Raymarching
Technologies
RustTauri v2Next.js 15React 19TypeScriptONNX RuntimeSilero VADSepFormerThree.jsTailwind CSS
Project Gallery
