Looking for a job

Soham Jadhav

@sohamjadhav

AI Engineer at ML4Sci developing CAC segmentation pipelines; benchmarked U-Net at 0.9416 Dice and ~63x faster inference.

India

Message

At Machine Learning for Science (ML4Sci), I'm developing reproducible CAC segmentation benchmarks on the Stanford COCA dataset. I benchmarked U-Net against TotalSegmentator, reaching a 0.9416 Dice score and ~63x faster inference on targeted CAC diagnostic workflows.

I built Convo-Ease, a multimodal moderation framework for text, images, and audio that was published in Cureus, Springer Nature. My work also includes PrediCT Studio, an auditable coronary calcium scoring workstation, and open-source contributions to pgmpy and pyaptamer.

Experience

Work history, roles, and key accomplishments

Current

Research & Development

Current

Machine Learning for Science (ML4Sci)

May 2026 - Present (5 months)

Developing reproducible CAC segmentation benchmarking pipelines on the Stanford COCA dataset to evaluate deep learning architectures for clinical-scale cardiovascular imaging. Benchmarked U-Net against TotalSegmentator, achieving a 0.9416 Dice score and ~63x faster inference efficiency.

Deep Learning U Net TotalSegmentator Benchmarking Medical Imaging Python PyTorch

Co-Lead | Artificial Intelligence & Machine Learning

Google Developer Groups (GDG) On Campus, Nashik

Sep 2024 - Sep 2025 (1 year)

Led the AI & ML vertical, architecting and delivering 10+ hands-on technical workshops on Deep Learning, RAG, and LangChain orchestration for an audience of 80+ engineering students. Provided direct mentorship on PyTorch and HuggingFace workflows.

Deep Learning RAG Lang Chain PyTorch Hugging Face Mentorship Workshop Delivery