Sounak Saha

I am a computer vision and machine learning engineer focused on building robust, real-time AI systems for research and robotics. I combine practical model development with deployment skills to deliver end-to-end solutions.

During an ML research internship at JNU, I developed a CNN-based real-time emotion recognition system with OpenCV and facial feature extraction, optimizing with batch normalization, dropout, and augmentation to reach 0.9746 accuracy.

At ERIC Robotics I created a YOLOv11-based foreign object detection model, deployed via Flask, and optimized the detection pipeline for constrained robotic environments, enabling seamless integration with robotic workflows.

I have built full-stack AI projects including a RAG-based research assistant (FastAPI, React, Pinecone), an automated loan approval app (Gradient Boosting, Streamlit), and a custom GPT from scratch (Transformer architecture), emphasizing reproducible deployments with Docker.