About

Learn more about me

Shamit Savant

ML Engineer · Graduate Researcher · Deep Learning & LLMs

I'm a Machine Learning Engineer and M.S. ECE graduate student at the University of Florida (graduating May 2026). Currently working as an ML Engineer Co-op at the UF Intelligent Clinical Care Center (IC3), building production-grade ML systems for medical image analysis at scale on HiPerGator HPC.

  • Email: savantshamit@gmail.com
  • City: Gainesville, Florida
  • Degree: M.S. ECE (May 2026), B.Tech EE
  • GPA: 4.00 / 4.00

My work spans production ML engineering, deep learning research, and LLM-powered systems. At IC3, I engineer large-scale image processing pipelines processing 200M+ pixel slides. Previously at PwC, I built GPT-4 extraction pipelines and AWS microservice infrastructure for enterprise clients. I've also co-authored an arXiv preprint on deep learning for medical image super-resolution.

Resume

Check My Resume

Download Resume

Summary

Shamit Savant

ML Engineer building production deep learning and LLM systems. Focused on medical imaging, computer vision, and scalable ML infrastructure.

  • Gainesville, Florida
  • savantshamit@gmail.com

Education

Master's in Electrical and Computer Engineering

Aug 2024 – May 2026

University of Florida

GPA: 4.00/4.00

Coursework: Deep Learning in Medical Image Analysis, Database Management Systems, Computational Photography

Bachelor of Technology in Electrical Engineering

2018 – 2022

Veermata Jijabai Technological Institute, Mumbai

GPA: 8.99/10.00

Coursework: Fundamentals of Machine Learning, Python Programming, Applied Linear Algebra, Probability and Statistics

Technical Skills

ML & Deep Learning

PyTorch TensorFlow HuggingFace Scikit-Learn OpenCV CNNs Vision Transformers Transfer Learning MONAI

LLMs & Agents

LangChain LangGraph OpenAI GPT-4/Vision RAG FAISS ChromaDB Pinecone Multi-Agent Systems

Infrastructure

Python C++ SQL CUDA PyTorch DDP HiPerGator HPC AWS ECS/EC2 Docker Git CI/CD

Professional Experience

UF Intelligent Clinical Care Center (IC3)

Jan 2026 – Present

Machine Learning Engineer Co-op, Gainesville, FL

  • Engineered a production image registration and feature extraction system processing 200M+ pixel images across an 8-slide cohort, extracting features across 820K+ cells per slide on HiPerGator HPC via affine registration, memory-aware coordinate transforms, and optimized rasterization
  • Reduced pipeline peak memory by ~70% on previously failing large-scale workloads by diagnosing three concurrent OOM failure modes and re-architecting data flow with adaptive downsampling and backward-compatible transform serialization
  • Recovered 1.6M annotations and 3+ hours of compute after a production job failure with zero reprocessing, by designing a checkpoint-and-replay mechanism from intermediate outputs already persisted to the server

Computational Microscopy Imaging Laboratory (CMIL), UF Medicine

Jan 2025 – Jan 2026

Research Assistant — Machine Learning, Gainesville, FL

  • Unblocked deep learning pipeline operations across 3 partner sites by diagnosing a GPU driver incompatibility introduced by a cluster upgrade, enabling the lab's full framework migration
  • Reduced team data operations overhead by 4+ hours/week by developing and open-sourcing a Python automation toolkit for bidirectional HPC-to-cloud transfer and annotation pipeline management, adopted as standard lab tooling on GitHub

PricewaterhouseCoopers (PwC)

Jul 2022 – Aug 2024

Technology Consultant, Mumbai, India

  • Reduced manual document processing time by ~75% for a client handling 12,000+ compliance documents monthly by engineering a GPT-4 extraction pipeline with domain-specific prompt engineering and custom chunking, achieving 91% field-extraction accuracy
  • Reduced deployment time by 50% and maintained p95 latency under 200ms through 200% peak traffic surges by migrating client services to containerized microservices on AWS ECS with ALB and auto-scaling, sustaining 99.9% uptime

Research & Publications

Prostate MRI Through-Plane Resolution Enhancement via Deep Learning

arXiv preprint, 2026
  • Benchmarked 5 deep learning architectures (CNN, U-Net, GAN, Attention-GAN, Diffusion) on 46K prostate MRI slices (TCIA, 58 patients)
  • Best model achieved 30.08 dB PSNR and 0.898 SSIM — 10.1% above the linear interpolation baseline; finding that problem formulation had greater impact than architecture choice

Projects

My Works

Featured

arXiv 2026 Prostate MRI Super-Resolution

Prostate MRI Super-Resolution

Deep learning benchmarking across 5 architectures on 46K MRI slices for through-plane resolution enhancement

Deep Learning Multimodal Vision-Language Model

Multimodal Vision-Language Model

Dual-stream Transformer with rotary embeddings — 18% alignment gain over CLIP baseline, 38GB → 26GB GPU memory reduction

LLM / RAG GPT-4 Vision Multimodal RAG Pipeline

GPT-4 Vision Multimodal RAG Pipeline

Hybrid retrieval system with 25% relevance gain over BM25 — 90%+ QA accuracy, query latency cut from ~1.5s to under 800ms

Handwritten Digit Detection

Handwritten Digit Detection — Faster R-CNN

Object detection system achieving 76.3% AP (IoU=0.50) on 8,000+ images using PyTorch on HiPerGator

Quadruped Robot

Quadruped Robot

Designed a mammal-inspired quadruped robot capable of multiple gaits and terrain traversal; mentored junior students through the Eklavya program

GreenHack Hackathon

Transmission Mast Detection — GreenHack Hackathon

Developed an RCNN model for identifying transmission mast types from drone imagery at an international hackathon

EV Powertrain Final Year Project

EV Powertrain — Modeling, Simulation & HIL Implementation

Final year project: modeled and simulated an EV powertrain with real-time Hardware-in-the-Loop (HIL) implementation

Street Light Fault Detection

Street Light Fault Detection — Smart India Hackathon

Led a team to build an automated street light monitoring system with GSM-based location tracking for service engineers

4x2 SRAM Array Design

4x2 SRAM Array Design

Design and simulation of a 4x2 SRAM array using 45nm technology in Cadence Virtuoso

Multi-Type CPU RISC-V Simulation

Multi-Type CPU RISC-V Simulation — Gem5

Simulating multi-type CPU processors using Gem5, focusing on RISC-V architectures and cache hierarchy analysis

Virtual Assistant

Virtual Assistant

Voice-controlled shell assistant — run directory, file, and navigation commands via speech

Diamond Grading

Diamond Grading — Deep Learning

Automated diamond grading system using Mask R-CNN, evaluating clarity and color from video frames

Contact

Contact Me

My Address

Gainesville, Florida, USA

Social Profiles

Email Me

savantshamit@gmail.com

Call Me

+1 (352) 709-6XXX