Rajath Koratagere Manjunath

Built domain agent skills for Claude and Cursor covering billing, legacy, and migration flows, with test accounts embedded per flow so agents could run end-to-end integration tests and validate their own changes. Designed and deployed an MCP server providing adaptive, context-scoped access control and log query, and led a 5-engineer team delivering a company-wide Zuora billing integration.

Built a 360° perception system stitching multiple cameras and integrated YOLOv7 on GPU for real-time obstacle detection. Eliminated a ~20% frame-drop rate in the AWS event-capture pipeline and cut P95 query latency from 1.8s to ~500ms by restructuring data and moving services off legacy bare-metal hosts.

Built an incentive recommendation model on SageMaker, trained on Redshift data and served from EC2 endpoints, forecasting site-specific hiring trends. Built end-to-end services managing bonuses and incentives for warehouse workers, processing ~$500K/week across a highly available AWS architecture.

Built training and testing pipelines for CV networks in PyTorch and Docker, and replaced manual tests with an automated CI pipeline on Jenkins. Developed C++ visual safety checks for Level 3 self-driving from CNN outputs, built map-data APIs, and a Qt real-time visualizer for vehicle state.

Designed an end-to-end neural motion planner for a 7-DoF arm, achieving 90% less planning time than RRT* at a 15% path-cost increase. Modified YOLOv3's loss for pedestrian detection and concurrently interned at D3 View modeling barrier impact and passenger motion in crash simulation.

Pursued a Master of Science in Electrical & Computer Engineering with an AI focus, from 2018 to 2020.

Earned a Bachelor of Engineering in Electronics & Communication from 2014 to 2018.

Completed an LLM-from-scratch cohort covering pretraining data, tokenizer pipelines, and transformer internals.