Venkatesh Mondi

Grace is a tool that generates an optimal model-serving stack with out-of-box monitoring and logging infrastructure. Users can deploy deep learning models that incur minimal technical debt by treating infrastructure as data without writing a line of code. I architected and led a team in developing Grace.

We leveraged open source projects (Docker, Keras, TensorFlow, TensorFlow Serving, PyTorch, ELK, Prometheus, Grafana, Flask, Nginx, and ONNX), which decreased the developmental time. We also set up automated load testing.

ADA analyzes automobile images to assess vehicle damage using object detection and classification. I modeled a deep learning pipeline to detect parts and classify damage, developed a custom tagging tool using surrogate models to reduce annotation effort, and identified essential images to be tagged using unsupervised analysis.

AVC is an AI-based system that classifies vehicle types from CCTV camera video footage. We achieved benchmark accuracy by using a novel approach to video classification, completed an on-premise deployment, and serviced India’s busiest toll gate processing 100,000 vehicles per day.