Six Agents, One Patient Record: A Google ADK Retrospective
- Category
- AI & ML
- Format
- Lightning talk — 15 minutes
- Audience level
- Intermediate
We hear a lot about how to build multi-agent systems, but much less about what gets harder once several agents have to work together. I built a healthcare AI workflow with six specialized agents using Google ADK. With six agents, the patient's context had to cross every boundary between them. In this lightning talk, I'll follow one synthetic patient record through a real execution trace and show where the handoffs introduced context-transfer overhead, latency, and new debugging challenges, and what I'd change if I built the system again.
Speakers
Radwan Baba
AI Engineer · Info-Tech Research Group
Radwan Baba is an AI Engineer at Info-Tech Research Group, where he designs and ships production AI and agentic systems. His work spans retrieval-augmented generation (RAG), multi-agent orchestration with LangGraph, Model Context Protocol (MCP) integrations, and enterprise AI on Azure OpenAI and Microsoft Fabric. He focuses on what makes LLM systems hold up in production: retrieval quality, agent orchestration, evaluation, observability, and cost and latency trade-offs. With a full-stack background, he treats AI as a systems design problem, not just a modelling one. Radwan holds a B.Eng. in Software Engineering from Concordia University and is pursuing an M.S. in Artificial Intelligence at Johns Hopkins University, focusing on clinical and healthcare applications of AI.