Rohan Chaudhari

– Developed a Crop Health system that calculates crop health indexes for up to 400,000 requests per day – Developed a framework called core-modules to easily scales geo-spatial and EO jobs on up to 100 instances. – Converting EO R&D pipelines to scale-able products. – Managed data traffic of up to 13 TB of data every 5 days from satellites. – Helped designing a library that takes care of scaling, infrastructure and abstracts complex processing of satellite data.

– Developed a pipeline for fetching of streaming data from micro-mobility provider APIs. – Helped setup CI/CD in the company and hence optimizing development times. – Developed a streaming metric calculation and a spatio-temporal aggregation system to help cities map out patterns in micro-mobility users. – Introduced data caching into the system to reduce latency of the platform. – Implemented a geo-zoning, street-association and other geographic aggregations in the application. – Database design and administration – Developed a robust back-end system that served 24000 spatio-temporal points (with metadata) per request.

– Was involved in research on how to deal with imbalanced image datasets and increase the accuracy of the models. Used various kinds of SMOTE to generate new data to balance the classes. – Developed Fire and Smoke detection models (Deep Learning based) which inferred based on video feeds from the camera. These models are currently being used to detect accidents in tunnels in France. – Worked on developing a solution to read License Plates from vehicles at toll systems. – Built pipelines to extract useful data from video streams and retrain new models based on that data. Used YOLOv2 and YOLOv3, Faster - RCNN. – Improved the process of retraining using automation of iterative and inter-dependent manual processes. (Significant speed-up) – Increased robustness of the pipeline to different European countries that vary significantly in physical appearances – Developed a model to detect vehicles from a camera stream from a drone. Used RetinaNet.