Spatial Sentinel

Inspiration

I have become very fascinated with Multi-Agent systems as of late and I also really like computer vision. I saw the track and ran through some ideas with Gemini, Claude, and some Google searches until I settled on this idea.

What it does

It uses agents with gemini models to analyze 3D spatial data and to help the user understand their environment better. best suited for smart glasses.

How we built it

Built and deployed the agents using ADK, Agent engine, and Github Codespaces, stored short term context using MongoDB MCP server, used model armor for input sanitization to protect from prompt injection.

Challenges we ran into

The initial code for the implementation didn't run in codespaces and after 3 hours of debugging it still didn't run. building and orchestrating the agents was challenging as well as findiing the right model and data for the job

Accomplishments that we're proud of

Built the multi agent system and secured it. got it working on video data

What we learned

What's next for Spatial Sentinel

getting it working with live video feed, adding performance monitoring to make it easier for human in the loop optimization, and adding more robust security for multimodal threats

Built With

• adk
• gcp
• python
• streamlit