Machine Learning Engineer at Infinium Robotics (2023-02 – Present)
- Built and maintained scalable ML pipelines for real-time box counting, handling data ingestion, model training, inference, and deployment in production environments.
- Developed a Pallet Segmentation model using MaskRCNN with custom modifications, enabling precise geometric measurements; deployed optimized model into production.
- Improved box detection accuracy by implementing a Boundary Guided Network on top of RetinaNet and integrating custom loss functions from research.
- Designed continuous data pipelines for dataset versioning and retraining, improving model robustness and performance.
- Optimized inference pipelines for latency and throughput in real-world warehouse environments.
Machine Learning Engineer at Kabam Robotics (2021-11 – 2023-01)
- Architected a real-time video analysis pipeline enabling distributed processing of multiple ML models using asynchronous thread pool execution.
- Designed and deployed scalable streaming pipelines using AWS (KVS, KDS), Docker, and Kubernetes (EKS), handling high-throughput video ingestion and inference workloads.
- Containerized ML services and explored Kubernetes-based orchestration for scaling inference workloads across multiple services.
- Developed spill detection system using DeeplabV3 (0.67 mAP) through model ensembling under limited data constraints.
- Built people counting and tracking system using Kalman Filters and YOLO with custom logic for static object detection.
- Integrated ML pipelines with robot mission planner using MQTT protocol for real-time communication.
Research Engineer Intern at Tieset (2021-07 – 2021-10)
- Developed Deep Q Learning-based dynamic pricing model to optimize hotel revenue.
- Built reinforcement learning model for autonomous driving (speed and braking control) using CARLA and OpenAI Gym; deployed on NVIDIA Jetson.
Machine Learning Intern at Dataeaze Systems (2020-05 – 2021-06)
- Developed active learning-based human-in-the-loop data labeling pipeline, improving annotation efficiency.
- Built car damage detection system for insurance automation and deployed using Streamlit.
- Optimized neural networks using TensorRT (quantization + pruning) for edge deployment on Jetson Nano.
- Developed MaskRCNN-based asbestos fiber detection system with preprocessing optimizations using Otsu's method.
