Build ML Infrastructure for Generative Recommendations

Tech lead of Instacart ML Training Platform

* Lead multi-disciplinary team to design, develop, and launch Instacart ML Training Platform to host production training jobs covering multiple ML frameworks (Tensorflow, PyTorch, XGBoost, LightGBM, CasualML etc) across all product lines (ads, search, recommendation and fulfillment).

* Introduce Ray as the primary distributed computation framework to Instacart ML Training Platform.

PyTorch in FB recommendation platform production: Built up PyTorch based distributed training framework SDK and ML consumer faced distributed training applications on both CPU and GPU platforms

• - Design & implementation of embedded LTE firmware algorithms for physical layer, with focus on downlink channel estimation & noise estimation; algorithm performance enhancement on multi-threaded systems for Qualcomm DSP Modem products.

• - Awarded 3 “Qual Star”s in recognition of contribution in latest modem firmware bring-up.

• - Worked with 4G LTE modem firmware development team. Implemented highly optimized embedded DSP firmware algorithms for physical layer on multi-threaded DSP systems. Emphasis is on architectural level optimization.

• - Adapted noise cancellation algorithms to multi-threaded modem architecture. Minimized execution MIPS by 80% while maintaining algorithm correctness on various field tests.