Haohan Zeng

About Haohan Zeng

Machine Learning Engineer specializing in LLM-based AI solutions, recommendation systems,…

Experience

1. Software Engineer

   Amazon Web Services (AWS)

   Jul 2022 — Present · Beijing, CN

   AI Agent for Cross-Border Logistics: Designed and implemented an AI agent leveraging Supervised Fine-Tuning (SFT), Mixture of Experts (MoE), and Reinforcement Learning from Human Feedback (RLHF) on AWS to enhance adaptability and decision-making. Optimized MoE gating mechanisms to reduce computational cost by 30%, fine-tuned a LLaMA 3-based model for logistics queries, and improved inference accuracy by 26% while reducing operational errors by 20%.Multimodal AI Chatbot for AWS Services: Developed a multimodal AI chatbot integrating SFT, CLIP, Stable Diffusion, and Retrieval-Augmented Generation (RAG) for AWS-related queries and image-based analysis. Fine-tuned a LLaMA 3 model for AWS-specific Q&A, enabled architecture diagram recognition with CLIP, and enhanced real-time retrieval using RAG, achieving a 35% accuracy improvement in troubleshooting and a 40% reduction in inference latency.

Education

• New York University

  Master's degree, Computer Science

  2018 — 2020

• University of California, Davis

  Bachelor's degree, Computational statistics, Computer Science, Psychology

  2014 — 2018