Andrew Mao

Machine Learning Engineer at DRW

Professional Experience

DRW

Software Developer | July 2022 - Present
Chicago, Illinois, United States · Hybrid

Generative AI

  • Built AI/ML/NLP systems for language model research, evaluation, search, retrieval, and deployment
  • Managed infrastructure for scaling ML workloads (data, inference) to multiple nodes and GPUs
  • Implemented retrieval augmented generation (RAG) pipelines, vector store indexing, agents
  • Built distributed data pipelines (Spark/Ray, ETLs), debugging tools for production services
  • Decreased model fine-tune/inference time and memory by 20% with tensor parallelism
  • Technologies: Python, PyTorch, Tensorflow, Ray, DeepSpeed, LangChain, MLFlow, Numpy, Spark, Pandas, Flask, PostgreSQL, Kubernetes, Javascript, GRPC/protobuf, Kafka, CI/CD

Automated Trading, Risk Analytics

  • Developed trading-critical systems for financial risk analytics, automated trading, price/size calculation
  • Managed microservices for distributed data processing (Kafka, Python GRPC APIs, ETLs, Kubernetes)
  • Worked on full product lifecycle from research (Jupyter notebooks) to productionizing (K8s), testing, and support
  • Developed tools for debugging/tracing system state, data flow, visualization, simulation, and scenario analysis
  • Implemented numerical/statistical models for option risk surface dynamics, PCA/matrix decomp, regressions
  • Technologies: Python, Pandas, Numpy/Scipy, Flask, Kubernetes, Pyarrow, GRPC, Redis, distributed systems

Meta

Software Engineer Intern | May 2022 - July 2022
Menlo Park, California, United States · On-site

  • Worked on the Language Understanding and Question Answering (LUQA) team, under Reality Labs AI
  • Optimized neural language models (LMs) for long-form question answering (QA) and summarization tasks
  • Fine-tuned and benchmarked retriever/reader models (BERT, RoBERTa, DPR) on meeting/conversation datasets
  • Researched model quantization, distillation, pruning, dense retrieval, attention models
  • Developed web apps to collect and crowdsource datasets on evaluating model generated output, demo capabilities
  • Technologies: Python, React, PyTorch, BERT, RoBERTa, DPR

University of Maryland

Research Assistant | June 2019 - May 2022
College Park, Maryland

  • Wrote and published papers in NLP conferences (EMNLP, EACL) on the topics of question answering, human-AI collaboration and topic models
  • Developed web app to accelerate coding (annotation) of technical documents, using active learning and NLP
  • Achieved theoretical speed up of 4x with comparable accuracy using topic models and text classification models
  • Applied computer vision techniques to multiple object detection, pose estimation and action recognition
  • Technologies: Python, PyTorch, NLP, Computer Vision, Web Development