Sarabjot Singh

Some places I have worked at,

• Research Intern @ RAFT Group, IIT Hyderabad

During my time at the RAFT Group, IIT Hyderabad, under the guidance of Dr. Satish Regonda, I worked on flood extent mapping and flood depth estimation using computer vision. I built and annotated datasets, benchmarked semantic segmentation models, and automated pipelines for object detection and keypoint tracking in flood imagery. I also explored semi-supervised learning to reduce manual labeling and improve model adaptability across diverse flood scenarios.

• Machine Learning Intern @ Stemtec AI and Robotics Pvt Ltd

During my time at Stemtec, under the guidance of Dr. Balakrishnan Ramalingam, I researched and implemented perception systems for autonomous mobile robots (AMRs), focusing on semantic segmentation and context-aware object detection. Additionally, I contributed to the development of a telepresence robot integrated with a large language model (LLM) for Human-Robot Interaction.

• Samsung Prism Research Intern @ Samsung R&D Institute India

During my time as a Project/Prism Intern at Samsung, under my academic mentor Dr. Mohanaprasad K, I worked on developing deep learning models for speech restoration and enhancement. My focus was on creating lightweight models designed for resource-constrained environments, enabling precise audio analysis and restoration, particularly for deployment on devices like smartphones.

My past projects

• Ollash Developed a natural language to shell command assistant using local LLMs with REPL and history search via Interro. Integrated support for Ollama and Hugging Face backends with a focus on safety and offline usability. Check out Ollash on PyPI: pip install ollash.
• MRIG - Medical Report Generation Developed MRIG using ResNet50 CNN architecture, enhancing it with additional neural network components for detailed chest X-ray analysis and interpretation. Engineered an algorithm to extract 13 critical parameters from chest X-ray scans, translating image features into actionable medical insights for healthcare diagnostics.
• VitaFile Led the frontend development of VitaFile, utilizing Google's APIs to centralize electronic health record (EHR) access, ensuring seamless integration and efficient data management. Introduced advanced features such as multilingual support and voice interaction, significantly enhancing the user experience and accessibility of VitaFile.

My Blogs

Work in progress, hopefully will be up soon.

Backlog of projects I want to build

• Decoding Semantic Segmentation I am working on the most comprehensive blog explaining semantic segmentation from first principles, covering traditional techniques and modern deep learning architectures with their key contributions, differences, and interpretability through research papers and implementations.
• Image Inpainting Tool Designed and developed an image inpainting system inspired by tools like Adobe Photoshop's Content-Aware Fill and Google's Magic Eraser. Initially implemented patch-based inpainting based on the paper "Context Encoders" , and later transitioned to DDPMs for high-fidelity object removal and semantic image editing.

Many more things too :)

This website was last updated on 18st August, 2025.