A summary is below. For a full resume, reach me at wanzelin007@gmail.com or on LinkedIn.
Experience
Senior AI Software Engineer, Postman (Apr 2026 to Present, San Francisco)
- DRI for model training and evaluation, owning the post-training pipeline and the evaluation environment that judges it.
- Build a multi-phase LLM fine-tuning pipeline on large-scale GPU infrastructure (SFT, DPO, RLVR with GRPO, and reinforcement fine-tuning), with data-quality filtering and benchmark-driven checkpoint evaluation.
- Design an industry-grade evaluation benchmark and environment for frontier LLMs with deterministic validators, bootstrap-CI scoring, anti-contamination, and multi-trial averaging, plus an abstention-adjusted metric separating capability from refusal.
- Design and deploy a next-generation agentic AI system and dynamic LLM routing strategies for a production enterprise platform.
- Primary technical point of contact for AI partnerships.
AI Research Scientist, Bobyard (May 2025 to Apr 2026, San Francisco)
- Built and shipped a production VQA and RAG system (FastAPI, PaddleOCR, LightRAG, AWS) over a two-layer knowledge graph with per-project data isolation.
- Architected a representation-learning pipeline for 100-megapixel and larger engineering drawings and the distributed GPU/CPU infrastructure behind it.
Graduate Research Assistant, Virginia Tech (May 2020 to Dec 2023, Falls Church, VA)
- Designed a decision-theory-guided deep RL framework for fast learning (up to 184% higher initial reward than standard deep RL) and an uncertainty-aware human-AI teaming system using Vision Transformers and Evidential Deep Learning, across four U.S. Army-funded projects.
- Published nine peer-reviewed papers on reinforcement learning, uncertainty quantification, and cyber-defense.
Graduate Teaching Assistant, Virginia Tech (Dec 2023 to May 2025)
- Taught Natural Language Processing (CS 5624), Decision Making Under Uncertainty (CS 5914), and Introduction to Artificial Intelligence (CS 4804).
Research Scientist Intern, Intelligent Fusion Technology (May 2022 to Aug 2022, Germantown, MD)
- Built deep RL prototypes (A2C, double DQN, dueling DQN, prioritized experience replay) and applied explainable AI on U.S. Air Force Research Laboratory funded projects.
Education
- Ph.D. in Computer Science, Virginia Tech, 2025. Dissertation: Game-Theoretic and ML-based Defensive Deception for Cyber-Physical Systems. Advisor: Dr. Jin-Hee Cho. Outstanding Ph.D. Research Award.
- M.S. in Computer Science, Virginia Tech, 2021.
- B.S. in Computer Science (Minor in Mathematics), University of Arizona, 2019.
Selected honors
- Outstanding Ph.D. Research Award, Virginia Tech, 2025.
- John Lee Pratt Endowed Fellowship in Computer Science, 2025.
- College of Engineering Graduate Student Publication Fellowship, 2023.
Technical skills
- LLM post-training: SFT, DPO, RLVR (GRPO), RFT, reward design, data curation and trajectory filtering, LoRA.
- LLM evaluation: agent and tool-use benchmark and environment design, anti-contamination, bootstrap-CI scoring, multi-trial averaging, capability-vs-abstention metrics.
- Engineering: Python, PyTorch, distributed GPU/CPU pipelines, Docker, AWS (EC2, S3, SQS), FastAPI, TypeScript/Node.js.
- RL and computer vision: classical deep RL (PPO, A3C, DQN), OpenAI Gym, PyBullet, computer vision (ViT, SAM, YOLO, OpenCV).
Service
Reviewer for IEEE TDSC, IEEE TIFS, IEEE Internet of Things Journal, IEEE TNSM, and others. Section Editor, Computer Software and Media Applications. Program Committee, IEEE TPS 2024.
Publications
15 papers (9 first-author), with 4 under review. See the Publications page or Google Scholar.
Languages
English (professional), Mandarin Chinese (native), Spanish (basic).