• Built an end-to-end action recognition pipeline that could be trained with as little as 1 hour of video data.

• Worked on data labeling pipelines to get data labeled using AWS Sagemaker GroundTruth and MTurk.

• Worked on converting FP32 Tensorflow models to FP16 TensorRT engines, which helped reduce the inference time and decreased the memory footprint of the overall ML pipeline by about 20%.

• Working on object detection model to detect region of interest in images for a visual inspection system.

Graduate Teaching Assitant for Machine Learning Course at Georgia Tech.