Pallab Maji

Developed and deployed solutions using NVIDIA Technologies for Retail Analytics, Recommendation Systems, Multi-lingual NLU, Autonomous Vehicles, and Industrial IoT applications. Expertise includes working with deep architectures (RCNN, FPN, SSD, UNET), training models with AMP & QAT on multinode GPU servers, optimizing with TensorRT, and deploying on NVIDIA Embedded Platform (DRIVE & Jetson). Experience in designing and developing AI datacentres with multiGPU clusters and deploying cluster management with Kubernetes & KubeFlow.

Led projects including Neural Network based Pedestrian and Cyclists Gesture Recognition and Scene Understanding. Developed common feature extractors using Inception, Resnet, FPN, and Mobilenet architectures. Worked with object detection architectures like RCNN, SSD, and YOLO. Optimized various networks for NVIDIA Drive platform using TensorRT. Developed ROS based integration for in-vehicle testing.

Developed computer vision based automatic website crawler using Robotic Process Automation. Developed algorithms for ranking clickable regions using Bayesian Network Model and performed website page segmentation using convolutional neural network.

Developed algorithms to detect road boundaries/edges. Used computationally cheap texture features (DCT, Color spaces, HOG, LBP) for semantic segmentation. Worked on semantic segmentation for scene understanding using deep learning algorithms like CNN, fine-tuning pre-trained models like FCN and SEGNET.

Developed algorithms for Intelligent Control for plasma confinement in Tokamak Fusion Test Reactor. Implemented algorithms in FPGA and DSP processor using Verilog and C. Learned fuzzy parameters from control data using Neural Networks and genetic algorithm.

Thesis – On Design and Implementation of Generic Fuzzy Logic Controller

Thesis – Application of Residue Arithmetic in Communication and Signal Processing