Guau S.O.S - Agentic Lost Pet Finder

An autonomous AI agent network powered by Gemini 2.5 Pro & Vertex AI to rescue lost pets in real time and lay the blueprint for future civil safety.

Comment 2

Inspiration

Every pet owner's worst nightmare is losing a member of their family. Traditional search methods—like posting static flyers or sharing on social media—are slow, passive, and geographically inefficient. We wanted to transform this desperate waiting time into an active, community-powered search network. Furthermore, we realized that the underlying framework of geolocated multi-modal AI agents could serve as a powerful proof-of-concept for the future of civil safety, such as rescuing missing children and lost individuals during humanitarian emergencies.

What it does

Guau S.O.S is a dynamic, collaborative civil-safety ecosystem:

  1. Active Map Dashboard: Pet owners create a "Lost Alert" and track the community feed in real time on an interactive Leaflet map featuring dynamic escape path estimations connecting successive chronological sightings.
  2. Guau Matcher Agent: When a community member sights a dog and uploads a photo, our background AI agent autonomously activates. It filters active cases geographically within a 20 km radius and performs a deep visual biometric analysis to compare both photos.
  3. Autonomous Actions: If the matching probability exceeds 70%, the agent calls our custom notifications tool to trigger high-priority push alerts (via FCM) to the owner's phone—even if it is locked.
  4. Real-time Live Chat: A private chat instantly opens between the witness and the owner to safely coordinate the pet's return.

How we built it

We engineered a highly robust, multi-tenant serverless stack:

  • Frontend: Next.js 16 (App Router), TypeScript, and TailwindCSS for maximum responsiveness.
  • Database & ORM: Prisma ORM with optimized SQLite database indexes to guarantee instant load times and fast query processing.
  • Backend Cloud Agents: Deployed via Firebase Cloud Functions (v2), Firestore triggers, and Google Cloud App Default Credentials (ADC).
  • AI Core: Google Cloud Vertex AI (Gemini 2.5 Pro Vision) with native Function Calling to enable structured autonomous decision-making (notifyOwnerOfMatch vs discardSighting).
  • Real-time Alerts: Firebase Cloud Messaging (FCM) configured with priority channels (guau_sos_channel_v3) for native sound triggers on iOS and Android.

Challenges we faced

Migrating from raw API keys to Vertex AI in enterprise-grade multi-tenant hosting environments presented security challenges. Ensuring empty referrer requests from serverless edge functions passed Google's strict authentication constraints required a robust transition to GCP Service Accounts. Optimizing image processing with Gemini 2.5 Pro Vision while maintaining tight latency constraints was another major hurdle, solved by applying Haversine-based pre-filtering on database coordinates before calling the multimodal model.

What we learned

We learned the incredible power of Function Calling inside Gemini. By giving the model access to specific tools like notification dispatchers, we transformed a simple "analytical AI" into an "active autonomous agent" capable of making real-world decisions safely and reliably.

What's next for Guau S.O.S

Our vision goes far beyond pets. The core architecture of Guau S.O.S is fully scalable. We plan to adapt this proven blueprint of decentralized geofencing, real-time community networking, and autonomous visual matching to partner with civil safety organizations to locate missing children and vulnerable individuals, turning a pet-saving app into a life-saving human safety network.

Built With

Share this project:

Updates