Self-taught developer since the pandemic. I build libraries that are as fast as I can make them, improving as I learn.
Everything is Python at the surface, but I write CUDA or C underneath when Python alone isn't fast enough.
Currently building a B2B personalized products company. I keep the business-specific code private, but share the general-purpose libraries I create along the way.
PhotoFF GPU-Accelerated
High-performance CUDA image processing library. Real-time filters, blending, and compositing - all on GPU.
- Order-of-magnitude faster than PIL/Pillow (see benchmarks)
- Smart memory management - allocate once, reuse efficiently
- Zero GPU-CPU copies - everything stays on GPU
- Originally built for custom OBS-style rendering engine
⚡ Next generation in development: Pure Python.h API (no CFFI), FFmpeg integration, DirectX/OpenGL interop without GPU→CPU copies
Transform any Python function into a complete web app automatically. Just type hints and run().
- 454 tests passing - battle-tested reliability
- Zero configuration - working web app in 30 seconds
- Auto-generated forms, file uploads, dark mode, validation, plots, downloads
from func_to_web import run
def divide(a: int, b: int):
return a / b
run(divide) # → http://127.0.0.1:8000📖 Full Documentation • PyPI Package
Turn Python functions into desktop GUI applications. The original version that inspired FuncToWeb.
- Cross-platform (Windows, macOS, Linux)
- Real-time updates or manual execution mode
- Built on Kivy
from functogui import App
def is_even(number: int = 4) -> bool:
return number % 2 == 0
App(is_even)Simple Python server deployment toolkit. Deploy to remote servers with just a few lines of code.
- Deploy FastAPI/Flask apps to $5 VPS servers
- Built on Fabric - mix with pure Fabric commands
- Supervisor service management
Fast G-code bounding box generator and framing tool in pure C.
- 26x faster than Python
- Zero runtime dependencies
- Handles 500MB+ files in seconds
- Universal compatibility (CNC, laser, 3D printer)
Fast Python logger with optional Telegram notifications. Separate channels for routine logs and critical errors.
- 195k logs/sec (Linux) / 80k logs/sec (Windows)
- Thread-safe with crash protection
Simple library for controlling CNC machines with GRBL firmware.
- Intelligent buffer management
- Auto-recovery from alarms
- Auto-disconnect detection
Pull requests welcome. I appreciate clear issues, minimal examples, and focused PRs.