Ranxin Li

About

Built and deployed Clyra.AI, an AI-driven daily scheduling product that uses a LangGraph-based multi-agent LLM pipeline (task extraction, verification, reflection) grounded with strict RAG over emails/documents/calendars and real-world signals like health metrics. Designed a custom agent orchestrator with bounded loops/termination conditions and a self-auditing verification/reflection layer to reduce hallucinations while controlling latency and cost via caching and model distillation.

Key Strengths

• Led end-to-end build and production deployment of an LLM multi-agent scheduling system (Clyra.AI)
• Designs reliable agent workflows using structured state objects (vs free-form text)
• Strong hallucination defenses via strict RAG + verification + reflection (self-auditing pipeline)
• Builds custom agent orchestrators with stateful graphs, conditional edges, and bounded loops to prevent runaway execution
• Balances quality vs cost/latency using embedding caching, selective agent invocation, and model distillation
• Measurable evaluation mindset (F1 for routing/classification; end-to-end latency metrics; hold-out testing)
• Translates non-technical product goals into concrete AI behaviors using scenarios and failure cases

Experience

Education

Skills

Languages