-
-
Chat with ADK
-
Architecture Diagram of AegisAI
-
Graph for Agent Usage and Allocation
Inspiration
It all started with a moment we’ve all faced — feeling unwell and turning to the internet for answers. We were met with a flood of confusing, often terrifying information. That’s when it hit us: why isn’t there a single, friendly assistant that truly helps people with their health concerns — both physical and mental? From late-night Googling to real-world struggles in finding trusted healthcare support, we were inspired to create AegisAI — a smart, empathetic digital agent that makes wellness more accessible, intuitive, and human.
What it does
AegisAI is like a caring friend with a brain full of medical wisdom. It listens to your symptoms and gives you preliminary diagnosis support. It offers mental health counseling, addiction recovery aid, dietary and fitness coaching, and more. It uses Google’s GenAI and Places API to help you locate nearby doctors or resources tailored to your needs. It even understands context and emotion, thanks to Gemini Flash on Vertex AI. All in one sleek chat interface — available on web, no app required.
How we built it
We built AegisAI using Google's Agent Development Kit (ADK) — which allowed us to craft modular, intelligent sub-agents, each focusing on specific areas like fitness, mental health, diagnosis, etc. Our stack included Python, Vertex AI, and Gemini 2.0 Flash. The UI was integrated directly with ADK using adk web for seamless testing and interaction. We containerized the project with Docker, hosted it on Google Container Registry, and deployed it on Cloud Run, making it scalable and accessible. And yes, we handled our environment variables securely using .env files for APIs and configurations.
Challenges we ran into
WSL2 refused to cooperate — Docker kept breaking on us until we reinstalled and configured everything from scratch. We battled with deployment errors like “PORT not exposed,” only to find the smallest misconfigurations. At one point, our AI kept hitting 404s — and it turned out the word “uscehtral1” was a typo for “us-central1”. That one cost us hours of head-scratching 😅. Managing .env placement and syncing configurations across local and cloud environments wasn’t as trivial as it sounds.
Accomplishments that we're proud of
We built and deployed a fully working AI agent from scratch! We integrated multiple APIs and sub-agents, giving AegisAI a holistic understanding of user needs. We transformed a local ADK prototype into a globally accessible Cloud Run service. Most importantly, we stayed resilient through every error, bug, and deployment failure — and made something that could genuinely help people.
What we learned
We gained hands-on experience with ADK, LLM-powered agents, and the GenAI ecosystem. We learned how to debug Docker builds and Cloud Run deployments like pros. We understood the real-world challenges of making AI not just functional, but useful and user-friendly. We learned that no matter how advanced the tech, empathy and user experience are what truly matter.
What's next for AegisAI
We plan to build a mobile-friendly version using Flutter or React Native. We’re looking to integrate secure user authentication, allowing for personalized wellness journeys. We’d love to bring multi-language support, making AegisAI accessible to people in rural areas and non-English speakers. And we want to collaborate with actual healthcare professionals to validate and enhance our AI's advice. AegisAI is not just a hackathon project — it’s a step toward human-centered AI in healthcare.
Log in or sign up for Devpost to join the conversation.