ADITYA KEERTI

At Peerprep, I built a marksheet extraction pipeline using OpenCV, YOLOv8, and table detection, then integrated OCR to handle different layouts. I also built a React frontend and async FastAPI/MySQL backend that processes a marksheet in around 30 seconds. In DOC-OC-v2, I replaced a multi-stage OCR and computer vision cascade with a local VLM service, bringing inference time to about 6 seconds per page. I’ve also built a clinical records application and the reinforcement learning component of a credit-limit optimization platform.