Deep Learning Engineer • Computer Vision Researcher • Applied Scientist
I work across the full ML stack — model design, training, large-scale data pipelines, and high-performance deployment. My primary focus is applied deep learning and computer vision, especially building systems that operate reliably in production.
I enjoy converting state-of-the-art research into robust, optimized systems used in real products.
- Vision Transformers, MoE architectures, DINOv3, SigLIP, YOLOv8-SegX, EfficientNet
- Multi-camera retail perception, SKU detection, segmentation pipelines
- Retrieval-based OOD detection, custom GPU kernels, flash-attention optimizations
- Training workflows: PyTorch, PyTorch-Lightning, FastAI, TensorFlow
- Triton Inference Server (Python backends and custom CUDA ops)
- High-throughput data loading, preprocessing, and augmentation pipelines
- Azure ML Studio, Docker, GitHub Actions, DVC, LanceDB, FiftyOne
-
Contributed to HuggingFace Transformers
- Improvements around ViT-MoE (issues/PRs #40145, #40215) and model behavior
-
Active in computer vision and training-efficiency communities
- Deep learning engineer at RadiusAI
- Previously co-founded/worked at a Berkeley SkyDeck–incubated startup
- Interested in large-scale vision systems, inference optimization, and agentic workflows
- Passionate about ML theory, mathematics, and building simple systems that scale
LinkedIn: https://www.linkedin.com/in/prajwal-anagani-a8401a191/ Website: https://lawjarp.is-a.dev/