Kishan Kumar

• Architected a centralized logging platform using Loki and Apache Ozone, processing 65 TB/day. • Resolved a critical bug in Loki 2.x, reducing storage footprint from 700 TB to 210 TB. • Created Gatekeeper for self-serve config management, reducing onboarding time from 48 hours to under 5 minutes. • Removed event-loop blocking by offloading I/O-intensive Aerospike operations to Vert.x workers; increased throughput to shrink fleet from 200 to 25 instances (-87.5%), saving 2,700+ CPU cores and 2.8 TB RAM

• Successfully automated fraud detection using Deep Learning techniques. • Trained a state-of-the-art RetinaNet model on a custom dataset, resulting in an accuracy of 98% for device classification. • Designed an innovative architecture, inspired by the human brain, utilizing RetinaNet as the backend for computing similarity between objects. • Implemented a three-stage similarity detection process, utilizing device IMEI, pHash, and extracted SIFT features.