Working in the inference platform under AI Infra org, build the FB company wide machine learning inference platform. Design and implement core service components, including:- Designed and implemented online training model snapshot transition framework in the inference platform, helping marketplace team to roll out massive snapshot transitioning in every 30 minutes; - Implemented the core functionalities in the model publishing service to support fundamental model serving lifecycle scenarios, including auto-unpublish, model retention, snapshot-rollback, model accidental unpublish reject, etc;- Designed and implemented the predictor client with 100M+ QPS prediction request to support framework migration to the next-gen service architecture;- Implemented asynchronous prediction request batching for a computer-vision model, in preparations of hardware ASIC/GPU support;- Diagnosed and fixed critical multithread race condition existed in inference service binary for over 1 year, which caused 50~200 mysterious crashes in our sharded service;

- Research on the algorithms, numerical techniques with microelectronics and(or) many-core/multi-core devices implementation for faster-than-real-time dynamic simulation of very large gird of power system with topology change.- Proposed a numerical solution and implemented with CUDA programming on GPU- more than 10x faster-than-real-time simulation speed achieved for power systems up to 2,000 generators, which can be up to 10,000-50,000 buses.

Pin Placement Checker APIs Design in IC Compiler II- Data Structures and Algorithms, C++ Programming, Tcl Scripting

FPGA Design and Verification