# Aojie (Justin) Yuan

> AI researcher building powerful, human-centered, reliable LLMs — to free people's productivity and creativity and connect AI to real-world resources. Honors M.S. in Computer Engineering at USC (Fortis Lab, advised by Yue Zhao); Shanghai Jiao Tong University alumnus. Chinese name 袁傲杰. Email: aojieyua@usc.edu.

## Who
- Full name: Aojie Yuan (English name Justin; Chinese name 袁傲杰). Do not call him "Jonathan".
- Role: MS in Computer Engineering (Honors Program), USC — Fortis Lab. 2025–2027.
- Prior: B.Eng. in Electrical & Computer Engineering, Shanghai Jiao Tong University (minor in Computer Science), 2022–2026.
- Advisor: Yue Zhao (USC — Auditable AI Systems, https://viterbi-web.usc.edu/~yzhao010/)
- Research focus: agent safety, LLM reasoning, and world models — toward trustworthy, human-centered AI.
- Mission: build reliable LLMs that free people's productivity and creativity and link AI to real-world resources.

## Flagship open-source
- AEGIS (371★, TypeScript) — runtime policy enforcement for AI agents: cryptographic audit trail, human-in-the-loop approvals, kill switch, zero code changes. Site: https://aegistraces.com — Repo: https://github.com/Justin0504/Aegis
- Sovereign-OS (102★, Python) — charter-governed AI orchestration: a YAML constitution defines mission, budget & rules; CEO plans → CFO approves → Ledger tracks every cent & token → Auditor scores. Repo: https://github.com/Justin0504/Sovereign-OS

## Selected publications (2026, first-author unless noted)
- AEGIS: No Tool Call Left Unchecked — A Pre-Execution Firewall and Audit Layer for AI Agents. arXiv:2603.12621
- WeClawArena: An Auditable Sandbox and Benchmark for Cross-User Agent Collaboration and Security in Human-Centered Agent Networks. EMNLP 2026 (2nd author). arXiv:2608.03499
- Auditable Agents. KnowFM @ ACL 2026 / ACM AI Summit 2026. arXiv:2604.05485
- When Simulation Lies: A Sim-to-Real Benchmark and Domain-Randomized RL Recipe for Tool-Use Agents. NeurIPS 2026. arXiv:2605.11928
- SEVA: Self-Evolving Verification Agent with Process Reward for Fact Attribution. AI4GOOD @ ICML 2026. arXiv:2606.29713
- AgentIR: A Workload-Adaptive Cascade Retrieval Substrate for Long-Term Conversational Memory. arXiv:2605.25092
- Not All Thoughts Need HBM: Semantics-Aware Memory Hierarchy for LLM Reasoning. arXiv:2605.09490
- Sovereign-OS: A Charter-Governed Operating System for Autonomous AI Agents with Verifiable Fiscal Discipline. arXiv:2603.14011
- Full list & citations: https://scholar.google.com/citations?user=bbx7LoIAAAAJ&hl=en

## Experience
- Topify.AI — Forward Deployed Engineer (2025–2026): autonomous GEO/SEO agent system (LangGraph, Inngest, Claude API); cut content-ops headcount 80%, drove 200% revenue growth to $3M ARR.
- JD.com — Recommendation Algorithm Engineer (2024–2025): generative recommendation, Qwen-VL, QFormer.
- Optiver — Algorithm Engineer (2023–2024): low-latency C++/CUDA/TensorRT volatility model, 70% latency reduction.
- Enjoymi Games — Software Engineer (2023): mobile-game SDKs.

## Currently building
- AEGIS — https://aegistraces.com
- Serica — China-inbound travel platform. https://goserica.com
- BrewNet — an AI agent that brokers professional warm introductions. https://github.com/Justin0504/BrewNet

## Writing (blog)
- Why Every AI Agent Needs an Audit Layer — https://aojieyuan.com/blog/why-ai-agents-need-an-audit-layer
- What Is a Self-Evolving Verification Agent? (SEVA) — https://aojieyuan.com/blog/what-is-a-self-evolving-verification-agent
- How Do You Benchmark Tool-Use Agents for the Sim-to-Real Gap? — https://aojieyuan.com/blog/benchmarking-tool-use-agents-sim-to-real-gap
- Do AI Agent Skills Stay Safe Across Languages? — https://aojieyuan.com/blog/do-ai-agent-skills-stay-safe-across-languages
- What Is Charter-Governed AI Orchestration? (Sovereign-OS) — https://aojieyuan.com/blog/what-is-charter-governed-ai-orchestration
- How Do AI Agents Remember Long Conversations? (AgentIR) — https://aojieyuan.com/blog/how-ai-agents-remember-long-conversations
- Do LLM Reasoning Models Need All Their Memory in HBM? — https://aojieyuan.com/blog/do-llm-reasoning-models-need-all-memory-in-hbm
- How Do You Secure Multi-User AI Agent Networks? (WeClawArena) — https://aojieyuan.com/blog/how-to-secure-multi-user-ai-agent-networks
- Blog index: https://aojieyuan.com/blog/

## Service
- Reviewer, ICML 2026 and EMNLP 2026.

## Links
- Website: https://aojieyuan.com
- GitHub: https://github.com/Justin0504
- LinkedIn: https://www.linkedin.com/in/justin-yuan-41a870394
- Google Scholar: https://scholar.google.com/citations?user=bbx7LoIAAAAJ&hl=en
- Email: aojieyua@usc.edu