Michael Zhao

AI/ML engineer with 6+ years of experience building production ML systems and LLM-powered agents at AWS and in frontier AI research settings. Designed multi-agent research workflows, RAG pipelines, and hybrid retrieval systems serving 2,000+ daily use with 97% user satisfaction. Deep expertise in agent orchestration, fine-tuning, ML system design, and AWS cloud infrastructure (SageMaker, Lambda, Step Functions)

Independent AI Research Scientist/Engineer at Independent AI Research Scientist/Engineer (2025-03 – Present)

• Design and developed research agent for StepFun support internal doc check, database search, code search, fetch document, find expert, verify claim to support research and development resulting in 2000+ daily usages
• Developed data agent auto parsing data from multiple sources such as pdf, JSON, Lark according to user question resulting in context sufficient rate increase from 80% to 90% +
• Applied intention detection, query enhancement paired with hybrid search including BM25 and embedding to route user queries on internal knowledgebase increased Recall@5 from 70% to 90%+
• Developed dynamic context management agents make sure agents get right amount of context in memory and store memory for long term if needed resulting in respond correctness increase from 80% to 90%+
• Applied multi-tiring safety control on agent behavior at run time and agent capabilities made sure agent run in safe context and only run in correct permission and instructions resulting in 98% task finish correct rate with 0 agent safety issue
• Created multi-agent workflow architecture plan complex jobs, using roles on demand and use proper tools, reflect and verify answers makes sure answer is correct and meet user intention resulting in 97% user satisfaction

Software Engineer, Artificial Intelligence and Machine Learning at Amazon Web Service (2021-07 – 2025-03)

• Designed and developed machine learning system to automate revenue forecasting streamlining the revenue planning process and resulting in 100% customer satisfaction while reducing planning time by 40%+
• Designed and developed a domain-specific GenAI chatbot with language model fine tuning and RAG to support new user onboarding, reducing onboarding time by 50%+ and cutting customer support tickets by 60%+
• Designed and implemented a full-stack solution for calculating 10M+ data records in financial planning, achieving over 100%+ performance improvement and 100% user satisfaction
• Designed and built a fully automated foreign tax rate allocation service for tax management users, reducing tax rate planning time by 50%+ and improving tax allocation forecast accuracy by 20%+
• Redesigned APIs, data models and architecture to make sure service is idempotent, improve readability, reusability, and maintainability, receiving high praise from both management and the team

Data Engineer at Joblogic-X Corporation (2019-08 – 2020-08)

• Developed ad-hoc report of inventory data with advanced ETL Techniques resulting in an efficiency increase by 80%
• Reviewed and troubleshot complex SQL queries for 20+ Utility & Gas related reports including Aging Report, No Standard Meter (NSM), Reduced Rate 175 and Switch Hold Report, resulting in data accuracy improved
• Collected data from multiple data sources such as data warehouse, HTML script, web log with SQL, Spark Python and regular expressions to develop predictive model with machine learning techniques resulting in 400k+ saving

Master of Science in Computer Science - Artificial Intelligence – University of Texas at Austin (2026-12)

Master of Science in Business Analytics – Southern Methodist University, Cox School of Business (2021-05)

Bachelor of Science in Accounting – University of Texas at Dallas (2019-12)