I’m a Lead Engineer building product systems where full-stack software meets AI infrastructure. My work spans developer tooling, local model runtimes, GPU-accelerated media, and the practical engineering required to ship them.

• Product engineering — durable web and desktop applications, from interface architecture through backend services and deployment.
• Applied AI — local inference, model training, agent workflows, and tools that make advanced models useful outside a notebook.
• Accelerated media — GPU-first image and video pipelines built around measurable output, explicit runtime contracts, and real hardware constraints.

LANGUAGES
C# · Python · TypeScript · JavaScript · SQL

APPLICATION SYSTEMS
.NET · ASP.NET · Angular · React · Node.js · Tauri · Electron

AI + GPU
PyTorch · CUDA · OpenCV · ONNX Runtime · FFmpeg · ComfyUI · local LLM tooling

DATA + DELIVERY
SQL Server · SQLite · Azure · Docker · GitHub Actions

Native NVIDIA DLSS processing for ComfyUI: frame interpolation, video upscaling, and image upscaling exposed as composable nodes. The project connects GPU runtime integration to practical video workflows, including deterministic frame-rate conversion, encoding, audio retention, diagnostics, and experimental Linux support through Wine.

Python · NVIDIA DLSS · ComfyUI · FFmpeg · GPU video

A compact GPT-style language model built from scratch in PyTorch, covering tokenizer training, pretraining, supervised fine-tuning, evaluation, and local inference. Its architecture explores modern small-model ideas including grouped-query attention, RoPE, RMSNorm, and sparse mixture-of-experts routing.

PyTorch · Transformers · BPE · CUDA · Model training

A local desktop-control agent for Ollama vision models. It turns natural-language tasks into a bounded observe → plan → act → verify loop, pairing a Tauri + React interface with a Python worker, validated actions, screenshots, and an MCP surface.

Python · Tauri · React · TypeScript · Ollama · MCP

INPUT   ambiguous product and systems problems
METHOD  model the contract · isolate the bottleneck · measure the result
OUTPUT  maintainable software that works on real machines

I’m most interested in work that crosses boundaries: UI and runtime, model and product, managed code and native acceleration. The through-line is practical systems thinking—understand the whole path, then make the critical path better.

Project work is the primary telemetry here. Browse the repositories, implementation notes, validation evidence, and active experiments directly on GitHub.