Gemini Ethos: Marathon Agent for Responsible Tourism

An autonomous AI agent powered by Gemini 3 that patrols protected ecosystems, analyzes tourist interactions in real-time, and ensures environmental ethics with transparency.

Comment

Inspiration

Fragile ecosystems are suffering from "over-tourism." Human rangers can't be everywhere 24/7. We were inspired to create a Marathon Agent that acts as a tireless, objective guardian to protect biodiversity through advanced AI oversight.

What it does

Gemini Ethos is a long-running autonomous agent that:

  • Performs real-time visual analysis of tourist-nature interactions.
  • Identifies species and human behaviors with high precision.
  • Generates Thought Signatures™, explaining its reasoning process and acknowledging uncertainties.
  • Features a Self-Correction Engine that recalibrates its assessment logic during long missions.

How I built it

  • AI Core: Integrated gemini-3-flash-preview via Google Vertex AI for multimodal reasoning.
  • Backend: Java 17 with SparkJava for a high-performance API.
  • Infrastructure: Deployed on Google Cloud Run for global scalability and 600s request timeouts.
  • Logic: Custom System Instructions designed for environmental law and causal analysis.

Challenges I ran into

The biggest challenge was handling high-resolution image processing without hitting network timeouts. We optimized the Java transport layer and implemented custom net properties to align with Cloud Run’s extended execution windows, ensuring the agent never "sleeps" during a mission.

Accomplishments that I'm proud of

  • Autonomous Reasoning: Successfully implementing Thought Signatures™, allowing the AI to show its "inner monologue" and logic steps.
  • Mission Continuity: Building a system that doesn't just analyze once, but maintains the context of an environmental patrol over long durations.
  • Technical Resilience: Overcoming complex Java/Cloud Run timeout challenges to ensure a stable 99.9% uptime for the agent's monitoring tasks.

What I learned

  • Agentic Design: How to move beyond simple prompts into a "Marathon Agent" architecture that manages its own state and corrects its previous errors.
  • Gemini 3 Multimodality: The incredible precision of Gemini 3 Flash in identifying specific South American species (like Alpacas and Tití monkeys) compared to older models.
  • Cloud Scale: Designing for high-performance AI workloads using Google Cloud Run and Vertex AI global endpoints.

What's next for Gemini Ethos

  • Drone Integration: Connecting the agent to live drone feeds for automated aerial patrolling of vast protected areas.
  • Multi-Agent Collaboration: Enabling multiple "Guardians" to communicate and share data about suspicious behaviors across different park sectors.
  • Local Language Support: Implementing real-time audio alerts in local languages to educate tourists immediately upon detecting a risk.

Built With

Share this project:

Updates