Skip to content
View AnnaSuSu's full-sized avatar
🤖
Building AI agents
🤖
Building AI agents
  • ShenZhen China

Highlights

  • Pro

Block or report AnnaSuSu

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
AnnaSuSu/README.md

Hi, I'm Aari 👋

Agent runtime & memory engineer · 78 merged PRs into agent-infra OSS · MSc AI · Shenzhen, China

I work on the layer under the agent — the harness, the sandbox, the memory, the parts that break in production.

🚀 About me

  • 🎓 MSc student in AI (2025–2028) — research focus: Agent Memory Systems
  • 🔧 Core external contributor to bytedance/deer-flow, a long-horizon SuperAgent harness
  • 🛠️ Creator of TechSpar — a memory-driven adaptive interview-coach agent (828+ ⭐, 108 forks)
  • 🌱 Open to AI agent / LLM infra internships

📦 Open source

78 PRs merged upstream since June 2026all of them here.

Most land in bytedance/deer-flow (71), where I work on harness internals: sandbox isolation and secret containment, MCP tool routing and validation, prompt-injection guardrails on every path where untrusted text reaches the model, checkpoint/resume correctness under streaming, long-term memory writeback, and the durable task runtime.

Elsewhere — Canner/WrenAI (6) · mem0ai/mem0 · infiniflow/ragflow and langgenius/dify in review.

🧰 Tech stack

Python FastAPI LangGraph Docker

  • Agent harness — multi-agent & subagents · skills · sandboxed tool execution · MCP · durable tasks · checkpoint/resume
  • Agent memory — Mem0 · multi-tier memory (STM/MTM/LPM) · context compression · conflict-resolution writeback
  • Retrieval / RAG — Graph-RAG / HippoRAG · hybrid retrieval + reranker · ANN indexes (HNSW/IVF) · RAGAS
  • Agent security — prompt-injection guardrails · untrusted-content escaping · skill/tool supply-chain scanning · secret isolation
  • Models / serving — SFT · LoRA · vLLM

🌟 Featured

  • TechSpar ⭐828 · 108 forks — memory-driven adaptive interview coach: a unified candidate profile drives training / resume / JD-prep / real-time copilot / review in one self-improving loop (LangGraph + 3-tier memory)
  • arxiv-mcp — MCP server for arXiv · EvoGraph · MemRepo

🏆 Honors

  • 📜 Patent — Microkernel-architecture context manager for large language models
  • 🥉 Kaggle Bronze — Jigsaw Agile Community Rules Classification (SFT track)
  • 🏅 2nd Prize — ASC25 Student Supercomputer Challenge
  • 🏅 3rd Prize — China Postgraduate Mathematical Contest in Modeling (2025)

📫 Contact · 📧 ansusu1025@gmail.com · 🔬 Kaggle

Pinned Loading

  1. TechSpar TechSpar Public

    一个持续进化的面试训练系统。在线体验地址:https://techspar.cn

    TypeScript 1.1k 126

  2. bytedance/deer-flow bytedance/deer-flow Public

    An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of…

    Python 81.2k 11.2k