Ke Xu 许可
Algorithm Engineer II · Huawei 2012 Laboratory
Summary
Algorithm engineer working on large-scale RLHF post-training and autonomous agent systems. At Huawei 2012 Laboratory I implemented GRPO/DAPO training modules on Ascend NPU clusters and now build multi-agent schedulers and recursive self-improvement (RSI) loops with held-out evaluation and auditable skill libraries.
Experience
Algorithm Engineer II · Huawei 2012 LaboratoryMay 2025 – Present
Large-Scale RLHF Training · Agent Systems for Scientific Discovery
- Implemented core modules of GRPO and DAPO for large-scale RLHF training on Ascend NPU clusters.
- Led GRPO training and tuning for large models, including dynamic sequence packing for long chain-of-thought samples and cluster-scale stability validation.
- Designed and built a pluggable DAG-based multi-agent scheduler with configurable node types and dependencies for a scientific-discovery agent framework.
- Building an RSI loop: rollout, failure-driven skill induction and top-K frontier selection over versioned, auditable skill libraries under strict train/validation/test discipline.
- Exploring open-ended evolution with island populations, MAP-Elites quality-diversity archives and tree search; next, self-generated evaluation tasks.
Open source RSI · AgentDescent / FlyBot
- Core contributor to AgentDescent, a parallel, asynchronous self-evolution framework where prompts, skills and code are versioned artifacts; candidates are evaluated on held-out tasks and only evidence-backed changes are committed.
- Work spans conflict-aware merging of parallel edits, acceptance gates, RSI metrics and auditable skill libraries.
- Contributor to FlyBot, RSI-oriented agent software: role-scoped bots split engineering work down a tree, with risk-tiered CI, review and a merge queue; it is the long-lived software counterpart to the AgentDescent framework.
Algorithm Engineer Intern · Ant GroupJun 2023 – Sep 2023
- Built a financial agent prototype on AntGLM and LangChain with ReAct-style reasoning, hybrid BM25/embedding retrieval and multi-turn tool calling.
- Fine-tuned AntGLM with LoRA for anti-money-laundering anomaly detection.
AI Engineer Intern · State StreetNov 2022 – Feb 2023
- Designed reconciliation algorithms for an AI-based automatic reconciliation system.
Research
Active Learning on Knowledge Graph Question AnsweringSummer 2022
Advisor: Prof. Hanghang Tong · UIUC · NSF REU
- Designed an active-learning strategy for KGQA fusing graph centrality, information density and node uncertainty.
Exploration of Reflective ASMs for SecurityApr 2021 – Apr 2022
Advisor: Prof. Klaus-Dieter Schewe · Zhejiang University
- Investigated reflective Abstract State Machines (ASMs) for security applications.
AI-Assisted Psychological Counseling ChatbotSummer 2021
Advisor: Dr. Zhenzhong Lan · Westlake University
- Built blacklist/whitelist safety classification and knowledge-graph-based intent and emotion recognition for counseling dialogue.
Education
M.S. in Computer Science · UC San DiegoSep 2023 – Dec 2024
Machine Learning concentration
B.S. in Electrical Engineering · UIUCSep 2021 – May 2023
Minor in Computer Science · dual-degree program with Zhejiang University
B.Eng. in Electrical Engineering and Automation · Zhejiang UniversitySep 2019 – Jul 2023
Honors
- 2012 Star Award, Huawei — 2026
- STDT Outstanding Project Delivery Award, Ascend Basic Software — 2026
- NSF REU Scholarship, National Science Foundation — 2022
- MCM/ICM Meritorious Winner (Top 7% of 10,053 teams) — 2021
Selected Skills
Python · PyTorch · RLHF · GRPO / DAPO · distributed training · Ascend NPU · LLM agents · self-evolving agents · multi-agent scheduling · RAG · LangChain · evaluation design · technical writing