DMS solutions. In-cabin sensing. Object recognition.Libraries/tools: Pytorch Lightning, Torchmetrics, Hydra, MlFlowOwner of face visibility and occlusion detection technology.

Edge ML. Experience with CNNs, LLMs, object detection, motion detection, object recognition, super-resolution. Experience with design, training and deployment of deep neural networks for edge devices. Model compression tools for deployment of different deep learning computer vision models on custom hardware. Owner for various ML projects. In charge with development of deep learning models, debugging and troubleshooting, kicking off experiments. Involved with all aspects of an ML application: data, data preprocessing, augmentations, model architecture, loss function, inference, evaluation, demos, experimentation on GPU clusters and debugging.Projects and areas I worked on:- Large language models: encoder, decoder models. ex. Roberta, LLama, GPT2Debugged evaluation pipelines for LLMs. Created demos.- Efficient Attention Mechanisms. Techniques to improve memory efficiency of LLMs.- Knowledge distillation in computer vision and LLM contexts.- Parallel computing. ex. DistributedDataparallel, Model sharding strategies for LLMs.- SuperResolution: ex. ESRGAN, Unet-type architectures. Design of evatuation tools.- Object detection : Yolov5, SqueezeDet, EfficientDet, M2Det - Network development and evaluation pipelines. - Kalman filtering. Object tracking.- Object recognition: Resnet50, MobilenetV2, SqueezeNet- Image segmentation, Optical Flow- Change Detection network for video surveillance. - Solved technical issues across mulitple projects that require mathematical skills and deep understanding of machine learning concepts.- Experience with classic ML and data science concepts gained during debugging and evaluation of models: feature selection, decision trees. Familiar with Jupyter notebook, numpy, pandas, sklearn libraries etc.Programming languages, frameworks and tools:- Python, Pytorch, Pytorch Lightning, Huggingface, Keras, Bitbucket, Jira

FPGA design