This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend I built WildGuard AI, an AI-powered multi-agent wildlife identification and safety assistant.
I built it for a friend who enjoys spending time outdoors, where encountering unfamiliar wildlife can quickly become a safety concern. The goal was to create something they could use when they encounter an animal they don't recognize.
A user can upload a wildlife image, and WildGuard AI coordinates multiple specialized agents to:
Instead of relying on one general-purpose agent, WildGuard divides the problem into specialized agents coordinated by an Orchestrator.
Live frontend:
https://satyakamspc.github.io/Wildguard-AI/ The project is currently deployed using GitHub Pages.
WildGuard AI — GitHub Repository WildGuard AI is built around Google ADK (Agent Development Kit) as the agent framework, with Gemini providing the model intelligence.
The system uses a multi-agent architecture consisting of:
The application is built with:
The project also uses dedicated Agent Skills for the AI/ML, backend, frontend, and database parts of the application.
WildGuard AI uses the open-source Google ADK as the foundation for its multi-agent architecture.
This was important because wildlife analysis is not a single task. Identification, geographic verification, risk assessment, first aid, and ecological education are different responsibilities that benefit from being separated into independently understandable agents.
Using an open agent framework allowed me to structure the application around specialized, inspectable components rather than putting the entire workflow inside a single opaque AI prompt. The model layer is separated from the agent architecture as well. Gemini currently provides the model intelligence, while ADK handles the agent orchestration and workflow.
This separation makes the system easier to extend: individual agents can be modified, tested, or replaced without redesigning the entire application.