• Leading Amazon’s model processing platform to optimize models in post-training for best inference performance

• Research and Integrate the cutting edge Inference technologies (Speculative decoding, Flash attention,

Quantization, Page attention) to achieve among between latency, throughput and cost

• Design and implement tools and pipelines for model optimization and deployment, speed up iteration by 5X

• Lead and integrate Hugging-face, Triton, FasterTransformer and TRT-LLM to production

Developing great AI platforms to make machine learning easier than ever for gas&oil companies, health industry, and finance industry

implemented auto-tune algorithm for hyper-parameter searching into training pipeline

* Evaluating machine learning models and rule-based model for lithography simulation (Large Geometry Data)

* Constructed pipeline using python, bash and Tcl to accelerate and automate data processing, resulting in a 80% decrease in time needed for evaluating software and model quality

* Increased accessibility and usability of customer data by redesigning data visualization techniques to include statistical graphs and information

* Improved object-oriented Python codes to check manufacturing parameters and reduced labor errors by 90%

* Developed new components in object-oriented Tcl codes and maintain codes with version control tools