Desktop applications and developer tools built with a local-first philosophy, prioritizing hardware acceleration, user privacy, and direct in-process execution.
SubtitleGo (Subtitle Studio)
v0.1.0Active ReleaseGPL-3.0
Standalone desktop AI speech recognition and subtitle editing studio. Runs the 1.7B Qwen3-ASR model entirely in local memory with CUDA, Metal (MPS), and CPU fallback. Features smart silence pause detection, clause pacing, live video player preview, and batch processing.
Key Capabilities
In-process Qwen3-ASR model execution with zero cloud or API dependency.
Silence-based pause segmentation with line character constraints.
Synchronized video preview player with interactive cue seek and edit.
Architecture & Target
PySide6 (Qt) frontend with multi-threaded background worker queues.
Tested on Windows 11, Linux (Ubuntu/Arch), and macOS Apple Silicon.
Standalone portable Windows package bundled with local FFmpeg binaries.
PySide6 (Qt)Qwen3-ASR 1.7BPyTorchCUDA / Metal (MPS)FFmpegVAD SegmentationSRT / WebVTTPython 3.10+
Deep reinforcement learning Go engine implementing the core AlphaZero architecture. Features a 19-block Residual Network (ResNet) with dual Policy and Value heads, 17-plane board state history with Ko rule handling, and batched PUCT Monte Carlo Tree Search with virtual loss. Engineered with a centralized GPU worker queue to bypass the Python GIL and maximize throughput without VRAM bottlenecks.
Key Capabilities
Dual Policy/Value ResNet predicting move probabilities and win evaluations.
Centralized GPU inference queue for multi-core CPU MCTS self-play without OOMs.
Interactive Pygame board interface with human vs. AI mode and pass mechanics.
Architecture & Target
PyTorch 2.x with CUDA 11/12 Tensor Core acceleration and CPU fallbacks.
Configurable board sizes (9x9 default, scalable to 13x13 and standard 19x19).
Full checkpoint saving/loading support with iteration snapshot replay.
A lightweight, self-hosted web application for team project management built with PHP 8.3 and MySQL 8.0. Features drag-and-drop Kanban boards, full task lifecycles with hierarchical subtasks, built-in time tracking with live timers, team role permissions, an in-app notification center, and REST JSON API endpoints.
Key Capabilities
Interactive Kanban boards with customizable sections and drag-and-drop ordering.
Active time tracking per task with pause/resume timers and report summaries.
Team management with role permissions (Admin, Manager, Member, Viewer).
Architecture & Target
PHP 8.3 front-controller with PDO prepared statements and session authentication.
One-command Docker Compose orchestration with automated database migrations.
Lightweight Vanilla JS frontend with Tailwind CSS styling and zero SPA framework bloat.
Feel free to reach out to me via email or connect on GitHub. Whether you want to discuss local AI architectures, collaborate on open-source software, or share ideas, I'd love to hear from you.
