Pet-First-Response

• 

  location_agent

• 

  initiate the search

• 

  web_post_agent

• 

  Agentic Workflow

• 

  traces and tool calls

Inspiration

When I lost my dog Barni in Upstate New York, the next fourteen days were a blur of frantic, manual labor. My world shrank to a checklist of phone calls to every shelter and vet, endless drives to facilities across county lines, and late nights posting on a dozen different websites and social media groups. Every task was a race against time, a repetitive and emotionally draining process that took me away from the most important thing: actually being out there, searching for him. After two relentless weeks, we were thankfully reunited, but I knew there had to be a way to automate the panic and give owners back precious time.

What it does

• AI “dispatcher” for lost pets: auto-runs search tasks the moment a case is filed.
• Predictive mapping: analyzes last-seen location to suggest likely routes & hiding spots.

• Multi-channel outreach:

  • TTS phone calls to nearby shelters.
  • Browser automation posts on lost-pet sites.
  • Flags key social-media groups for the owner.

• Owner deliverables: a ranked action plan showing automated steps and next on-the-ground moves.

How we built it

• Framework: Google Agent Development Kit (ADK).
• Orchestrator: Gemini-powered PetFinder Orchestrator coordinates the workflow.
• Sub-agents:

1. location_agent – geo prediction via Google Search.
2. voice_agent – TTS calls (Bland.ai/Play.ai-style).
3. browser_agent – site postings with Stagehand-like automation.
4. research_agent – compiles findings into the final report.
   • Architecture: modular, scalable, A2A-style communication using ADK’s AgentTool.

Challenges

• Sequencing mixed synchronous/asynchronous tasks (simple lookups vs. stateful browser flows & live calls).
• Preventing the orchestrator from ending early after a “task complete” signal.
• Fix: strict state management + explicit prompting to enforce full plan execution.

What we learned

• Agentic decomposition beats a single monolith for complex workflows.
• Crystal-clear prompts are vital for reliable inter-agent hand-offs and overall flow control.

What’s next

• Community plugins: one-click postings to Nextdoor, local forums, etc.
• Real-time sighting network: SMS/web app feeds live reports back into the location_agent for dynamic re-routing.
• Richer data sources: transit feeds, weather APIs, and other context to sharpen movement predictions.

Built With

• blandai
• browser-automation-(stagehand)
• browserbase
• gemini-2.5-pro
• google
• google-agent-development-kit-(adk)
• opentelemetry
• python
• stagehand
• text-to-speech-apis
• w&b-weave