๐ค AI/ML Engineer, building & shipping models ๐งฉ Author of KerasFormers & Trinix ๐ รใฐใค Kaggle Expert ๐ค Open-source contributor, Hugging Face & Keras ๐ Based in: Fullerton, USA
I'm an AI/ML engineer who builds open-source libraries that make modern models easy to use, from pretrained Keras 3 model collections to Triton-accelerated PyTorch layers. I work across computer vision, LLMs, RAG, and agentic AI, and I love turning research into clean, reusable code.
- ๐ฆ Author of KerasFormers, Keras 3 pretrained models across Vision, LLM, VLM, Depth & Speech
- โก Built Trinix, Triton-accelerated PyTorch layers with automatic fallback optimization
- ๐ รใฐใค Kaggle Expert
- ๐ค Contributor to Hugging Face (Transformers, Diffusers) and the Keras ecosystem (
keras-cv,keras-hub) - ๐ง Implemented SOTA vision models in Keras/TF: SegFormer, ConvNeXt, NextViT, YOLO
Pretrained Models for Keras 3
An open-source Keras 3 collection of pretrained models spanning Vision, LLM, VLM, Depth, and Speech, backend-agnostic across TensorFlow, JAX, and PyTorch.
๐ฅ Highlights:
- One API, multiple modalities
- Runs on TF / JAX / PyTorch backends
- Clean, reusable model implementations
โก Tech Stack:
Keras 3 JAX TensorFlow PyTorch
Triton-Accelerated PyTorch Layers
Drop-in PyTorch layers accelerated with custom Triton kernels, featuring automatic fallback optimization when a fast path isn't available.
๐ฅ Highlights:
- Custom fused Triton kernels
- Automatic fallback to native ops
- Plug-and-play with
torch.nn
โก Tech Stack:
Triton PyTorch CUDA
Keras Implementation of AdEMAMix
A faithful Keras / TensorFlow implementation of the AdEMAMix optimizer, a mixture of two EMAs for faster, more stable training.
๐ฅ Highlights:
- Paper-accurate optimizer port
- Drop-in
keras.optimizerscompatible - Reproducible training recipes
โก Tech Stack:
Keras TensorFlow
Run Hugging Face Models Locally
Browse the Hugging Face Hub, download with a click, and run detection, segmentation, depth, OCR, SAM, VLMs, ASR, TTS & diffusion fully locally. Your data never leaves your GPU.
๐ฅ Highlights:
- No cloud, no upload, 100% local
- Dozens of task types out of the box
- One-click model download & run
โก Tech Stack:
Hugging Face PyTorch Local Inference
Semantic Segmentation in TensorFlow
A clean TensorFlow / Keras implementation of SegFormer, efficient transformer-based semantic segmentation with hierarchical features.
๐ฅ Highlights:
- Transformer encoder + lightweight MLP decoder
- Reproducible training & inference
- Readable, well-structured code
โก Tech Stack:
TensorFlow Keras Transformers
Drive Coding CLIs from Your Phone
Control any coding CLI from your phone via WhatsApp, Telegram, Discord, Slack, Teams, or Signal, code from anywhere, no laptop required.
๐ฅ Highlights:
- Works with 6+ chat platforms
- Bridges chat โ your terminal
- Code on the go
โก Tech Stack:
TypeScript Node.js Chat APIs