Looking for a job

Ricky User

@rickyuser3

Message

I build production multi-agent AI platforms for healthcare and finance.

Hong Kong

Message

What I'm looking for

I'm seeking fully remote Agentic AI or LLM engineering roles where I can build production multi-agent systems, RAG pipelines, tool integrations, evaluation frameworks, and safety guardrails for real-world users.

At Topflight, I architected and led development of MD Genius, a four-agent healthcare platform deployed to two clinics and serving 100+ patients across six channels, including real-time voice. I built its shared tool, RAG, safety, persistence, and message-bus layers, reducing manual staff intervention by 80%.

Previously at Huawei, I built multi-agent financial and government-document LLM systems using hybrid retrieval, MCP tool calling, LoRA fine-tuning, and evaluation frameworks. My work reached 95% cold-start intent accuracy, was adopted by 20+ traders, and established an internal benchmark that improved domain writing quality by 50%+.

Experience

Work history, roles, and key accomplishments

AI Engineer

Topflight

Jan 2026 - Sep 2026 (8 months)

Core architect and lead developer of a multi-agent healthcare platform, owning six backend services and contributing about 60% of platform commits. Architected four specialized production agents with explicit handoff protocols and per-agent state machines, built a shared tool layer with 30+ idempotent tool definitions, and developed a production RAG pipeline with per-clinic isolation.

Multi Agent Systems Agent orchestration RAG Tool calling MCP Azure AI Search fastAPI PostgreSQL SQLAlchemy Azure Container Apps Azure Service Bus REDIS Streams Docker Prompt Engineering

Machine Learning Engineer

Huawei

Feb 2023 - Dec 2025 (2 years 10 months)

Core architect and lead developer of a multi-agent financial chatbot, splitting the assistant into specialist agents behind a router and standardizing external capabilities behind an MCP tool-calling interface. Built a dynamic RAG pipeline with hybrid retrieval and context-aware chunking, achieving 95% intent accuracy in cold-start sessions and adoption by 20+ traders.