SafeBite

landing page
user onboarding survey
dashboard
product scanning
scan results 1
scan results 2

Inspiration

Food labels can be confusing. Most people don’t really know what the ingredients on the back of a package mean, and we wanted to fix that. SafeBite started as a simple idea: help people understand what they’re eating and make better choices without stress.

We imagined an app that not only analyzes what’s inside your food but also learns your habits and shows them in a clear, beautiful way.

What it does

SafeBite lets you scan any food product and instantly tells you if it’s safe for you. It highlights sketchy or risky ingredients and, most importantly, explains why you might want to avoid them.

Everything starts when you upload an image to our app. SafeBite uses a multi-agent AI pipeline that analyzes the product, takes your preferences into account, and gives you a clear, personalized result you can actually understand.

How we built it

We built SafeBite as a mobile web app using React and JavaScript, with a clean and responsive UI designed using Tailwind CSS and Framer Motion for smooth animations.

For authentication, we used Auth0 to let users log in securely without the hassle of creating yet another account.

The backend runs on FastAPI (Python), which powers our REST endpoints and manages the multi-agent orchestration using the OpenAI Agents SDK. We also set up an MCP server to manage user preference data in a SQLite database through the mcp-memory-libsql package.

For identifying products from uploaded images, we integrated Google’s Gemini API, which uses multimodal LLM capabilities to extract product details with impressive accuracy.

Challenges we ran into

Getting the scanning experience to feel fast and smooth was tricky. We had to balance AI accuracy, speed, and clear communication between the agents without slowing things down.

Designing the Food Universe visualization also took some creative trial and error. It needed to look great while still being easy to interpret for users who just want to know, “Is this good for me or not?”

Accomplishments we’re proud of

We’re proud that SafeBite feels smart, clean, and genuinely useful. The multi-agent system actually works — it analyzes food, considers user preferences, and provides meaningful suggestions. That took plenty of experimentation (and API tokens).

We’re also proud of how the app looks and feels. The Food Universe visualization helps users see their eating patterns in a fresh, interactive way.

In short, we built a working app that connects good design with AI in a way that just makes sense.

What we learned

A lot! From building responsive UIs and orchestrating multiple AI agents to setting up MCP servers and managing data pipelines, this project helped us sharpen both our technical and teamwork skills.

The hackathon also reminded us of how valuable different perspectives are when it comes to building something people will actually use.