Jingyi Ni

I’m a Staff Software Engineer with 10+ years across Airbnb, Amazon, and Microsoft, where I’ve specialized in production multi-agent systems, RAG pipelines, and LLM infrastructure. I architect systems end-to-end and ship the code across backend, frontend, and infrastructure.

At Airbnb, I built Airbnb’s production agent runtime from scratch using a custom LangGraph-based stack—planning loops, dynamic tool selection, and long-term memory—then deployed it on AWS with multi-region failover, canary releases, and blue-green rollouts. I also led a 30+ tool ecosystem with schema validation, idempotency, and rollback gates, integrating model routing across Claude, OpenAI, and Llama via AWS Bedrock.

I design the safety and evaluation backbone for agents, including the SafeAgent Router and the Guardrails & Safety Engine for content moderation, prompt injection defense, PII redaction, and policy compliance scoring. I’ve also built evaluation harnesses with offline regression tests and LangSmith trace scoring to prevent regressions, and used RLHF-lite preference fine-tuning to improve task success.

Beyond core agent systems, I’ve shipped agent-facing experiences (like the Agent Console with streaming and human-in-the-loop workflows) and delivered platform improvements through reliability and performance engineering. I’m now looking to build from the ground up at an early-stage company where speed and ownership matter, and where mentoring and technical leadership can compound impact.