Founder & CEO @ FIM Labs · MS AI · Enterprise AI Agents & Document Intelligence · China × Global
Building cognitive architectures that give LLMs better memory than mine 🧠
class TaoAn:
def __init__(self):
self.location = "Singapore × Beijing"
self.company = "FIM Labs"
self.role = "Founder & CEO"
self.education = "MS in Artificial Intelligence, Hawaii Pacific University (2026)"
self.interests = ["RAG", "LLM Memory Architectures", "Knowledge Graphs", "Agentic Workflows"]
self.current_focus = "Enterprise AI agents & document intelligence across the China × Global boundary"
self.belief = "Most 'AI agents' should be workflows — fix the plumbing before buying a bigger pump."
def say_hi(self):
print("Thanks for dropping by! Let's build something cool together.")
me = TaoAn()
me.say_hi()
Product
- fim-ai/fim-one — open-source agent platform for Global × China enterprises (self-hosted, any LLM). Day-job / company product I build and ship.
Upstream (external)
openai/openai-agents-python— found a resume-time bug where a pending nested agent-tool approval could bind to the wrong tool call after an earlier entry was filtered out (#3749: analysis + patch + regression test). Maintainer re-landed a source-level fix as #3753 withCo-authored-by: Tao An.
Original research / tools
- cognitive-workspace — active memory / functional infinite-context architecture for LLMs.
- nano-spec — lightweight task-spec format for AI-assisted development.
- cog-canvas — training-free long-term memory for LLM conversations.