Beibei Cheng

About Beibei Cheng

10 years programmining experience developing machine learning/image processing algorithms/big data analysis.4 years professional software development experience shipping retail commercial titles.• Solid background in machine learning, data mining, statistical analysis, computational intelligence• Solid background in image processing, machine vision, digital signal processing, 3D image rendering and reconstruction • Platform familiarity: familiar with Xbox, Windows Smartphone, Embedded System Platform• Tool familiarity: familiar with R Programming, MATLAB, SAS, Minitab, Weka• Programming language proficiency: C/C++/C#, SQL, Python, ImageJ, OpenCV,HTML,XML, Java• General graphics experience (OpenGL, DirectX), heterogeneous compute (Multithreading, CUDA)• Engineer leadership in self-motivated, communication skill, and teamwork and passionate in working on fun side projects

Experience

1. Principal Applied Scientist

   Microsoft

   Mar 2025 — Present · Redmond, WA, US

   Entra Risk Management Agent. Designed and built an AI-driven identity risk investigation, triage, and remediation agent for large-scale enterprise tenants. Implemented long-term memory and user profiling so the agent could reason over behavioral history (not just point-in-time alerts),and correlated multiple risk signals in near real time to separate true positives vs. false positives with context-aware, explainable remediation recommendations. Fine-tuned GPT-4o-mini for identity risk investigation and remediation reasoning to improve consistency and explanation quality. Partnered with identity infrastructure teams to expose investigation/remediation actions as callable tools (MCP) for safe agent execution, and combined ML-based risk scoring with LLM-driven reasoning to deliver high accuracy with human-readable explanations. Built offline replay + automated grading to evaluate agent quality, prompt robustness, and data reliability pre-production, and designed feedback loops to improve downstream detection models.• Conditional Access Agent. Led the design and deployment of a multi-agent orchestration framework, automating the enforcement of Zero-Trust access policies across complex enterprise environments. Fine-tuned GPT-4o-mini for conditional access deep analysis and policy recommendation generation to improve decision quality and reduce prompt brittleness. Engineered large-scale Conditional Access Deep Analysis pipelines to fuse real-time identity telemetry with LLM decision logic, reducing manual analyst intervention by 43%. • Agent ID Protection. Applied anomaly detection models over high-dimensional security telemetry to safeguard agent identities. The system proactively detects and blocks threats by identifying anomalous activities that deviate from established behavioral baselines and Suspicious tool invocation sequences.

Education

• Missouri University of Science and Technology

  Ph. D, Computer Engineering

  2009 — 2012

• UESTC

  B.S., Electronic Engineering

• Missouri University of Science and Technology

  M.S., Computer Engineering

  2008 — 2009

Skills

• Matlab
• Python
• Algorithms
• Artificial Intelligence
• Digital Image Processing
• Neural Networks
• Programming
• Vlsi
• Embedded Systems
• Opencv
• Data Mining
• Signal Processing
• Pattern Recognition
• Computer Vision
• Machine Learning
• C++
• Computer Architecture
• Simulations
• Image Processing