shighram Kumar

I’m an AI Development Engineer focused on building practical machine learning and computer vision systems that can be deployed and tested in the real world. I’m currently pursuing an MTech in Robotics and Artificial Intelligence from IIT Bhubaneswar.

During my internships at Ranvox Technology, I worked on autonomous water trash detection and drone navigation without GPS. I tested YOLOv7n and MobileNetSSD on NVIDIA Jetson Nano, developed YOLOv11n for water trash detection, and used an OAK-D camera for face depth measurement and segmentation.

I also built a sea turtle detection and counting pipeline for Rushikulya Beach using SAM-3 auto-annotation and YOLOv8n training, along with a Python video inference system designed for accurate counting from video streams. In my current remote role, I developed a plank posture classification system with MediaPipe Pose and Logistic Regression, plus a rule-based squat counter for repetition counting and form validation.

I deliver solutions end-to-end, from model training to deployment on GCP for web-based access. My technical projects include semantic segmentation of kidney ultrasound images (U-Net variants and an EAMS-Net approach), production-level RAG/agentic AI chatbot pipelines, and a Tkinter-based sound classification app using Random Forest and PCA.