My Kaggle: https://www.kaggle.com/sajinpgupta •I created a 7 member deep learning team from scratch with much of early recruiting and structuring. I lead the computer vision team to develop video analytics based safety and surveillance product. The team is focused on all aspects of 1) data gathering (exploring public datasets, create image dataset, label images for safety equipment objects 2 ) Neural network based algorithms research and finalisation 3) Neural network training and evaluation for object detection 4) training monitoring and accuracy in improvement upto 95% 5 ) deploy the model and video pipeline in production. • I designed the heuristics data model for the video pipeline to support 1) Camera configuration 2) Model to Camera mapping 3) Model repository 4) Object detection results and labels storage by time • I am responsible for UI/UX team to design the product user interface screens for object detection results analysis on Timeline, events analytics, Camera and model configuration screens. • Created a 6 member Data Science team to build a Anomaly Detection and Failure Prediction Training and Inference Framework using Deep Temporal Clustering algorithm. This allowed us to early identify equipment failure detection as early as 4 to 6 weeks in advance. Reduce False Alarm Rate by 60% in production and reduced the Equipment downtime significantly by early detecting operation anomalies.Also training scalability of the framework grew to 100s of assets time series datasets • Built a product to automate defect detection in Manufactured components • Built an OCR product to extract numbers from license plates using CNN and LSTM networks with CTC loss. Achieved 95% accuracy with 10000 images in train and test • Built a face detection and recognition AI model to automate video analysis of CCTV surveillance footages • Built a classification model to predict state and plate type for a vehicle plate image while vehicle passing through the toll gate • Built a utility to locate number plate in large blurry vehicle plate images • Built an AI solution to automate failure detection in Solar PV panels. Failures are single hotspot, linear hotspot, panel failure, string failure and panel induced degradation •Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization, Structuring Machine Learning Project •Frameworks: Tensorflow, Torch, Keras, Python