cv

Software Engineer Intern

Instance Creator LLC

Developed advanced automation systems and semantic matching platforms for streamlining application workflows.

• Spearheaded a vector-based semantic matching system using Qdrant, boosting matching accuracy to over 80%.
• Developed a multi-agent automation platform with AutoGen and gpt-4o to orchestrate complex application workflows, reducing manual effort by 85%.
• Designed and implemented a scalable microservices backend using async FastAPI to support over 1,000 concurrent users; optimized performance by introducing a Redis caching layer, reducing external API latency and call frequency by 80%.

Software Engineer Intern

Epoching AI

Enhanced AI model performance and deployed scalable watermark removal services.

• Achieved a 10x inference speedup and 30% accuracy boost for a watermark removal service using NVIDIA TensorRT.
• Developed a data generation and augmentation pipeline, improving model robustness against semi-transparent watermarks.
• Containerized the service using Docker and deployed it on Cloud ECS, implementing Prometheus for real-time monitoring and ensuring high availability to handle 500+ concurrent requests.

Software Engineer Intern

Ytell Network Technology Co., Ltd.

Built high-performance multimodal systems and automated data processing pipelines.

• Developed a high-performance multimodal RAG chatbot with FastAPI, increasing inference throughput by 5x using Flash Attention 2 and cutting retrieval errors by 40% through a hybrid search strategy.
• Automated the processing of 1,000+ daily orders with a data pipeline using Apache Airflow and MongoDB, eliminating over 5 hours of weekly manual work for each store manager.

Master of Science

Carnegie Mellon University

Artificial Intelligence Engineering - Information Security

• LLM Systems
• LLM Methods and Applications
• AI Systems and Tool Chains
• Computer Systems

Bachelor of Engineering

University of International Business and Economics

Data Science and Big Data Technology

• Operation Systems
• Data Structure and Algorithm
• Natural Language Processing
• Big Data Analysis

ImaginAItion: AI Literacy Game

Engineered a real-time multiplayer AI literacy game with React (TypeScript) frontend and FastAPI backend for CMU HCI.

• Leveraged Socket.IO over WebSockets to ensure low-latency (<150ms) state synchronization for 50+ concurrent players.
• Implemented a stateful backend game loop to manage turn-based player progression, integrating the gpt-image-1 API for real-time, dynamic image generation.
• Orchestrated the full-stack application with Docker Compose for reproducible local development and deployed the containerized services to Amazon ECS for a scalable production environment.

Silent Supporter: Scalable AI Multimodal Therapy Platform

Built a scalable therapy platform on a Node.js backend for Tsinghua AIR with advanced AI capabilities.

• Reduced real-time generation latency by 80% via WebSockets; utilized PostgreSQL for primary data storage and Redis for session caching to improve performance.
• Achieved targeted music style transfer by fine-tuning the InspireMusic 1.5B model on a custom-built dataset, and developed a WebGL engine to dynamically visualize conversational emotion in real-time.
• Designed a low-latency, event-driven architecture using RabbitMQ for asynchronous, non-blocking multimodal generation; separately optimized the core AI model's inference speed by 3x through ONNX Runtime graph optimization.

MovieRec – End-to-End Recommendation System

Architected a full-stack recommendation platform with comprehensive MLOps workflow.

• Integrated a Vue.js frontend with a backend based on Flask API and SpringBoot; Leveraged MySQL and Redis to handle the data persistence layer.
• Trained Random Forest model using Scikit-learn, conducted offline evaluation with train-validation splits and online evaluation with telemetry data to assess model performance; managed A/B testing experiments with MLflow.
• Established a containerized MLOps workflow, simulating a high-availability architecture with Minikube and implementing a Jenkins CI/CD pipeline with Prometheus/Grafana for automated deployment and real-time monitoring.

Self-LLM: Open-Source LLM Deployment Guide

Co-authored and maintained a leading open-source guide simplifying LLM deployment with 23.2k GitHub stars.

• Led to feature in the keynote presentation at Google I/O Connect China 2024.
• As a core contributor, authored key tutorials on GLM4 deployment with vLLM, LangChain integration, and LoRA fine-tuning to simplify the learning curve for developers.