WorldCup26-Football-Travel-Agent

Agentic AI solution based on Google Cloud + Arize Phoenix Tracing/Observability to inform about World Cup 26 Football Soccer and facilitate the Search/Management of Travel between City Stadium Matches

Comment

Inspiration

Interested in Google Cloud, MCP and Agentic AI Technologies and as Football (Soccer) Fan User with upcoming World Cup 2026 starting from 11th of June, I intend to propose a project combining Google Cloud ADK+Gemini LLM ** and **Arize Phoenix Observability Technologies for this Google Cloud Rapid Hackathon.

What it does

The agentic application enables World Cup 26 Football (Soccer) Fan users to get information about the participating Teams, Groups, Players, Matches, and also to get Inspiration, Planning, Pre/Post Trip, Booking for Travels to reach the Matches Stadium Cities. In parallel, the application has been instrumented with OpenTelemetry Tracing to enable Administration Team to access Observability/Evaluation data on Arize Phoenix Cloud.

How we built it

  • World Cup 2026 Football (Soccer) Database about Participating Teams, Groups, Players, Matches has been introduced and designed as Relational SQL Database to facilitate the associated Football data searching. This Database can be loaded / managed/ updated with Administrative Python Scripts and is accessible as Football MCP service.

  • Agentic AI Google ADK Gemini/VertexAI LLM Application involve multiple dedicated sub task (Football, Planning, Pre/Post Trip, Booking) agents, under control of main orchestrator agent

  • Arize Phoenix Instrumentation has been integrated in order to to generate Observability Tracing Data in special Phoenix Cloud Space to enable Evaluation and later Improvement of the Agentic Application Behaviour / Performance. The associated Phoenix MCP service have been also integrated in the Agent Application. Running evaluation experiments with LLM-as-a-Judge as Python scripts has been launched

Challenges we ran into

  • Issue " Invalid input: expected record, received array" when interacting with Arize Phoenix MCP server @arizeai/phoenix-mcp on Gemini CLI (Windows) has required to notify Arize Phoenix (+Google?) support team. Problem has not been reproduced on MCP Inspector nor Claude Desktop tool

  • Cloud Run Deployment for Remote Access Issue has required to use a specific Dockerfile to fix it with also appropriate higher memory setting.

Accomplishments that we're proud of

  • Working Agentic AI Application with multiple Agents dedicated to Football / Travel (Planning, Pre-Post Trip, Booking) and also for Observability Tracing Evaluation

  • Optimized Football (Soccer) World Cup 26 Teams Database about Teams, Groups, Players, Matches which can be updated in Real Time to get Fresh Data

What we learned

  • Arize Phoenix Tracing / Observability / Evaluation integration, hosting (local, cloud)

What's next for MySearchAgent

  • Finalization of Prompts to improve the behaviour of Agents

  • Agentic AI (auto) improvement based on Phoenix Tool and Observability Data

Built With

Share this project:

Updates