Open-Source Tooling

Production-grade Python packages for reproducible ML research, edge deployment, and computer vision. All available on PyPI with MIT licensing.

ed-rvfl-sc

PyPI

MIT License · Python ≥3.9

GitHub

Ensemble Deep Random Vector Functional Link with skip connections. Delivers deep-ML–level accuracy without GPU training using closed-form layer solves and feature reuse. Targets ultra-fast training and inference on CPU and embedded boards (Jetson, Raspberry Pi).

CPU-only
No GPU required
~100×
Faster training setup
Edge-ready
Jetson · RPi · Arduino
pip install ed-rvfl-sc
Non-iterative training Skip connections Closed-form solve Ensemble learning

TimeMesh

PyPI

PyTorch · Temporal CNNs / LSTM / Transformer

GitHub

Preprocessing and dataset tooling for time-series forecasting experiments. Provides sliding windows, normalization flows, reproducible train/val/test slicing, and ready-to-train tensors for LSTM, TCN, Transformer, and iTransformer-style models. Designed for fast ablation studies and reproducible research.

pip install timemesh
Sliding windows Normalization Reproducible splits Multi-model support

Keras Swin-UNet

PyPI

TensorFlow · Vision Transformer · Focal Loss

GitHub

Transformer-based Swin-UNet segmentation stack for earth observation, medical imaging, and industrial perception. Includes attention backbones, skip connections, and robust Focal Loss settings for rare-structure segmentation. Clean Keras API compatible with TensorFlow.

pip install keras-swin-unet
Swin Transformer UNet decoder Focal Loss Earth observation