Aari AnnaSuSu

Hi, I'm Aari 👋

Agent Memory Systems researcher · building memory-driven AI agents · MSc AI · Shenzhen, China

I research how agents remember — and contribute to the open-source projects building it (cognee, mem0).

🎓 MSc student in AI (2025–2028) — research focus: Agent Memory Systems
🧠 I work on: giving agents durable long-term memory — tiered storage, knowledge-graph RAG, context compression, and agent harnesses
🛠️ Creator of TechSpar — a memory-driven adaptive interview-coach agent (700+ ⭐, 90+ forks)
🌱 Open to AI agent / LLM internships

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
Agents — multi-agent & subagents · tool use · planning · state persistence · MCP
Models / serving — SFT · LoRA · vLLM

Contributing to the agent-memory & agent-infra projects I research — PRs:

🧠 topoteretes/cognee (AI memory platform) — #3539 add CENTRALITY search type (PageRank/degree over the native graph projection)
💾 mem0ai/mem0 (agent memory layer) — #5910 fix Elasticsearch KNN result cap · #5908 lazy-import langchain-core in async procedural memory
🤖 langchain-ai/open-swe (coding agent) — #1628 stop agent factories mutating the caller's RunnableConfig
🦌 bytedance/deer-flow (SuperAgent harness) — #3812 fix MCP tool prefix-collision mis-routing
🏋️ OpenPipe/ART (agent RL trainer) — #741 fork checkpoints across W&B entities

TechSpar ⭐700+ · 90+ 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