WildGuard AI: A Multi-Agent Wildlife Safety Assistant

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend.

What I Built

I built WildGuard AI, a full-stack wildlife identification and safety assistant designed to help a friend—or anyone who enjoys hiking, traveling, farming, or exploring nature—make more informed decisions when encountering unfamiliar wildlife.

An unfamiliar animal can raise urgent questions: What species is it? Is it commonly found here? Is it dangerous? What should I do if someone is bitten or stung? WildGuard AI brings these questions into one workflow. Users upload a wildlife photograph and receive a structured report covering species identification, geographic plausibility, potential risks, precautionary and first-aid guidance, and ecological information.

The application includes image-quality validation, local database caching, and a React dashboard for viewing results. WildGuard AI is an informational assistant, not a substitute for professional species identification, emergency services, or medical advice.

Demo

Live demo: https://satyakamspc.github.io/Wildguard-AI/

Suggested walkthrough: upload a wildlife image, show the agent-driven analysis, and demonstrate the resulting safety report in the dashboard.

Code

WildGuard AI — GitHub Repository

How I Built It

WildGuard AI uses a multi-agent architecture rather than asking one general-purpose agent to handle the entire analysis. A central Orchestrator Agent coordinates six specialized agents:

1. Species Agent: Identifies the most likely species from an uploaded image.
2. Verification Agent: Checks whether the identification is geographically plausible using location information and available reference data.
3. Risk Agent: Assesses potential hazards and assigns an appropriate risk level.
4. First Aid Agent: Generates relevant emergency and precautionary guidance.
5. Knowledge Agent: Provides information about habitat, behavior, and ecological importance.
6. Report Agent: Combines the agents' outputs into a structured wildlife safety report.

The workflow begins when a user uploads an image through the React + Vite frontend. The Django backend receives it through a REST API, then starts the orchestrated analysis. The resulting report is returned to the frontend for display. SQLite supports local database caching, while Pydantic and Pillow are included in the Python stack.

The project uses Google ADK for its agent-oriented architecture, Google Antigravity 2.0 in the development workflow, and the Gemini API for AI-powered image analysis and agent capabilities. It also includes dedicated Agent Skills for backend, frontend, database, and AI/ML development.

Open-source AI disclosure: Google ADK is an open-source agent framework, but Gemini is accessed here through a proprietary API. I am not claiming that the current project runs an open-weight model or performs local inference. If the challenge requires an open-weight model specifically, that integration would need to be completed and documented before submission.

Why Does Open Innovation Matter?

Open-source agent frameworks make it easier to experiment with architectures that can be inspected, extended, and shared. For WildGuard AI, Google ADK provides a framework for organizing specialized agents and their orchestration, rather than locking the entire application into one monolithic prompt.

The separation between the Django application, agent responsibilities, and frontend also makes it easier to test components independently and explore alternative models or inference providers in the future. Community contributions could improve geographic reference data, safety-information review, accessibility, and support for additional species.

There is an important distinction: the framework is open-source, but the current Gemini inference dependency is not open-weight or locally hosted. An open-weight model could offer additional control over deployment, model inspection, and offline use. Those are opportunities for future development, not features I claim to have implemented already.

Prize Categories

• GitHub — Best Use of GitHub Copilot

  • Used GitHub Actions to automate the WildGuard AI frontend build and deployment workflow to GitHub Pages.

Built with: React, Vite, Django, Python, SQLite, Google ADK, Google Antigravity 2.0, Gemini API, Pydantic, and Pillow.