Hello! 👋

I'm Yiming Feng

AI/ML Engineer & Duke MSCS Student

About Me

I'm an AI/ML Engineer and Duke MSCS student specializing in LLMs, Agentic AI, and machine learning systems. I build end-to-end AI solutions, including RAG pipelines, multi-agent workflows, and deep learning models, with a focus on scalable, reliable, and production-ready applications.

Python C++ Java JavaScript / TypeScript SQL PyTorch Scikit-learn TensorFlow LLMs Agentic AI RAG Prompt Engineering Vector Databases Embedding Models Deep Learning CNNs Deep Reinforcement Learning Multi-Agent RL Self-Play Training Deep Monte Carlo Image Classification Object Detection (YOLO) OpenCV MediaPipe FastAPI Docker Kubernetes GitHub Actions Databricks Linux Git React Nuxt Node.js REST APIs WebSocket MongoDB Redis PostgreSQL

Education

Master of Science in Computer Science

Duke University

Aug 2025 - May 2027 (Expected)
  • GPA: 4.0

Bachelor of Science in Computer Science

University of California, Davis

Jul 2022 - Jun 2025
  • GPA: 3.9
  • Honors Graduation

Experience

AI/ML Intern

Qorvo

May 2026 - Present
  • Designed and developed an agentic GenAI workflow (As-Is Documentation Agent) that analyzes SAP source code and technical artifacts to automatically generate, summarize, and iteratively refine solution design documents and process documentation.
  • Implemented multi-stage LLM orchestration pipelines using prompt engineering, structured outputs, and automated validation loops to improve documentation accuracy, traceability, and consistency across enterprise business processes.
  • Built RECEFW Agent, a Retrieval-Augmented Generation (RAG) chatbot integrating multiple enterprise data sources, enabling context-aware question answering and knowledge retrieval for internal users.
  • Developed document ingestion, indexing, and retrieval pipelines utilizing vector databases, embedding models, and semantic search techniques to enhance response relevance and reduce hallucinations.

Graduate Teaching Assistant

Duke University CS Department

Sep 2025 - Present
  • Evaluate assignments promptly and provide detailed, actionable feedback that strengthens student understanding and performance.
  • Hold weekly office hours to answer questions and debug coding issues, improving comprehension of core course material.
  • Lead pre-exam review sessions with targeted practice problems to reinforce key concepts and improve exam readiness.

System Architect Intern

Guangdong Yixun Technology

Jul 2023 - Aug 2023
  • Integrated Vue.js with Spring Boot to streamline the user experience and improve system responsiveness.
  • Raised YOLOv3 image recognition accuracy by developing and curating specialized training datasets.
  • Built monitoring pipelines with Prometheus and Grafana to surface system health and performance metrics.

Projects

01

As-Is Documentation Agent

An agentic GenAI workflow that analyzes SAP source code and technical artifacts to automatically generate solution design documents, process documentation, and traceability matrices through a multi-stage LLM pipeline.

Agentic AILLMsDatabricksOpenAI SDK
02

RECEFW Agent

An enterprise Retrieval-Augmented Generation (RAG) chatbot integrating multiple internal knowledge sources with vector search and hybrid retrieval. Indexed 10,000+ enterprise documents to provide context-aware question answering with significantly improved retrieval accuracy and reduced hallucinations.

RAGLLMsVector SearchDatabricks
03

StyleDou

A style-conditioned DouDizhu RL agent built on DouZero/DouZero+, with opponent-style LSTM encoders and auxiliary next-action prediction. Achieved ~98% win rate vs. Random and ~83% vs. RLCard.

Reinforcement LearningOpponent ModelingDeep Monte Carlo
04

SafeBench-DRA

An offline Chinese–English benchmark for evaluating security failures in deep-research agents using deterministic retrieval, adversarial tool-response overlays, and trace-level scoring.

LLM AgentsDeep Research AgentBenchmarkingAgent Security
05

Hand Gestures Classifier

A multimodal deep learning system for classifying 18 hand gestures using Graph Neural Networks, ResNet-34, and late fusion. Achieved 99.72% accuracy on the HaGRID dataset.

GNNResNet-34Image ClassificationPyTorch
06

LOL-Win-Prediction

Predicted win probability in League of Legends matches using a deep neural network trained on first 15 minutes of in-game data. Improved local robustness with PGD attacks and TRADES.

Adversarial RobustnessPGDTRADESDeep Learning
07

Proactive-AI

A Chrome extension that detects user hovers and selections to surface relevant AI tools in place — enabling real-time code explanation, summarization, equation graphing, table extraction, and OCR.

Chrome ExtensionAI ToolsOCR
08

Chinese Handwritten Digit Classifier

Deep learning models with VGG16 and LeNet to classify handwritten Chinese numerals. Achieved ~97% accuracy on the Chinese MNIST dataset.

CNNVGG16LeNet
09

Blood-On-Mahjong

A real-time multiplayer Sichuan Mahjong web application with WebSocket-based gameplay and Google OAuth authentication, designed for scalable Kubernetes-based hosting.

TypeScriptWebSocketGoogle OAuthKubernetesCI/CD
010

WOAA-trading

A full-stack stock trading simulation platform with real-time market data streaming and historical backtesting, letting users practice and refine trading strategies without financial risk.

Full-StackWebSocketBacktesting