Mihir G.

Key Skills

Software

AWS SageMaker

Label Studio

OpenCV AI Kit (OAK)

Top Subject Matter

Neuroscience / Glioblastoma MRI tumor segmentation and forecasting

Robotics / autonomous navigation / aquatic environment trials

Top Data Types

Image

Video

Computer Code Programming

Top Task Types

Segmentation

Fine-tuning

Bounding Box

Classification

Computer Programming/Coding

Freelancer Overview

Student project: 3D pre-operative brain tumor volume modeling (U-Net, ConvLSTM). Brings 4+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include PyTorch. Education includes Bachelor of Science, Columbia University (2025). AI-training focus includes data types such as Medical, DICOM, and Computer Code and labeling workflows including Segmentation and Fine-tuning.

Labeling Experience

Research assistant project work: LSTM-driven intent estimation for obstacle avoidance in unmanned water vehicles

Fine-tuning

Developed an AI research workflow for autonomous navigation behavior using deep learning components and time-series intent estimation. Designed modeling approaches intended to support training and evaluation of perception-to-planning systems in dynamic environments. Integrated LSTM-driven intent estimation into obstacle avoidance strategies for unmanned water vehicles. • Built LSTM-based intent estimation components • Integrated learned intent into multi-objective navigation planning • Conducted research-oriented experimentation in aquatic trials with dynamic obstacles • Produced validated multi-modal motion planning techniques using sensor inputs

2024

Student project: 3D pre-operative brain tumor volume modeling (U-Net, ConvLSTM)

Segmentation

Built a multi-model computer-vision pipeline to segment and forecast brain tumor volume from patient MRI scans over a half-year horizon. Trained and optimized models to produce accurate tumor delineations and growth forecasts across time-series medical images. Used a GPU-optimized PyTorch workflow with a custom loss function to reduce training time significantly. • Performed semantic segmentation for 3D pre-operative brain tumor volume modeling • Implemented ConvLSTM-based temporal modeling for growth forecasting • Tuned training objectives via a custom loss function • Achieved reported segmentation/forecasting accuracy with fast inference

Not specified

Education

Work History