Inspiration
The 2026 World Cup will face unprecedented stadium logistics. Staff will be overwhelmed by walkie-talkie reports of crowd surges, medical emergencies, and broken turnstiles. I wanted to build an AI that doesn't just act as a chatbot, but acts as an Autonomous Director turning frantic natural language reports into structured, actionable database operations instantly.
What it does
Sentinel-1 acts as the lead operations manager for World Cup stadiums. When staff report an issue (e.g., "Massive water leak in Fan Zone 1"), the agent ingests the text, assesses the severity (Critical, High, Medium, Low), extracts the stadium zone, and autonomously triggers the MongoDB Model Context Protocol (MCP) tool. It automatically creates a dispatched operations ticket in the database and replies to the user with the action taken.
How we built it
I utilized the Gemini GenAI SDK (specifically Gemini 3 Flash for high-speed reasoning and precise tool-calling). Integrated FastMCP to define our tools and connected it to MongoDB using PyMongo. I designed a strict system prompt to govern how the agent assesses threats. Finally, wrapped the agent in a Streamlit user interface to provide a clean dashboard for stadium operators.
Challenges we ran into
The biggest challenge was bridging the gap between a conversational LLM and a strict database schema. I had to carefully tune the system instructions and the MCP tool parameters so that Gemini would reliably extract the correct zone and priority levels without human intervention. Additionally, building and deploying the system without Google Cloud credits required us to creatively utilize GitHub Codespaces and alternative hosting to meet the deadline.
Accomplishments that we're proud of
Successfully getting Gemini to step out of the "chat" box and autonomously execute database operations! Seeing the AI independently decide to trigger the log_stadium_incident tool and watching the structured JSON data instantly appear in our MongoDB cluster was an incredible "aha" moment.
What we learned
I learned exactly how powerful the Model Context Protocol (MCP) is. It essentially gives Gemini "hands" to manipulate the real world (or in this case, a digital twin of a stadium). I also learned that Gemini 3 Flash is exceptionally good at function-calling and adhering to strict operational protocols.
What's next for Sentinel-1: 2026 World Cup Operations Agent
I plan to expand the agent's MCP toolkit. While it currently manages the MongoDB operations database, I want to fully activate the GitLab MCP integration so that structural stadium failures automatically generate engineering pull-requests and dev tickets. I also aim to connect it to live stadium IoT sensors to trigger the agent without human input.
Log in or sign up for Devpost to join the conversation.