From Image to Itinerary: Multimodal Agentic Travel Planning with MCP, A2A, and BeeAI

​From Image to Itinerary: Multimodal Agentic Travel Planning with MCP, A2A, and BeeAI

​Abstract:

​Planning a trip from visual input is a deceptively complex problem for AI. In this session, we present a multimodal-first, local-first agentic architecture where a user uploads an image (e.g. “where is this place?”), and the system generates a travel plan using Model Context Protocol (MCP) and A2A (Agent-to-Agent), running fully locally without cloud dependencies.

​Models are executed using OpenVINO and OpenVINO Model Server, enabling optimized inference across CPUs and GPUs with minimal configuration. They are automatically optimized and exposed through simple local APIs, allowing developers to interact with them as if they were cloud services—without managing infrastructure.

​The system follows a router and specialist agent pattern, where dedicated agents handle image understanding, hotel search, and flight search via MCP servers. A multimodal model extracts meaning from the image, and the router decomposes and delegates tasks across agents through A2A.

​We focus on practical architecture, configuration, and lessons learned for building MCP-centric, multimodal systems that are extensible, reproducible, and production-ready—while remaining fully local and efficient.

About the speaker:

Ezequiel Lanza is an AI Software Engineer and Evangelist with a passion for making artificial intelligence practical and accessible to developers.

As a regular speaker at leading AI and open-source conferences, he creates impactful use cases, tutorials, and guides that help teams adopt and deploy open-source AI tools effectively.  With a strong engineering background and over two decades of experience supporting developers and customers in the software industry, Ezequiel brings a unique blend of technical expertise and real-world perspective. He holds a master's in data science at Universidad Austral in Argentina. 

Ezequiel was elected as a Board Member and TAC Chair of the Linux Foundation AI & Data, where he helped shape the future of open-source AI and foster collaboration across the global ecosystem.


Friday, May 1
Auckland - Wellington — 12:00 PM
Sydney - Melbourne — 10:00 AM
Singapore — 8:00 AM

​Thursday, April 30
Buenos Aires / São Paulo — 9:00 PM
Toronto — 8:00 PM
San Francisco — 5:00 PM