I Built an Open Source Health AI Agent Without a Vector DB (Laravel 12, React, Typescript + InteriaJS + Gemini)

Source: Dev.to

Introduction

Tossing a basic frontend on top of an AI chat API is something you can almost do in your sleep now. But building something that actually understands messy, personal health data—blood‑sugar numbers, specific calorie burns, real dietary limits—requires a different approach.

I open‑sourced Acara Plate, a platform for personalized nutrition. The goal was to move past generic “eat your vegetables” advice and create an assistant that works with a user’s unique body data, without getting buried in weeks of infrastructure management.

Bypassing a Vector Database

Traditional setups for health‑focused AI involve:

1. Chunking a massive dataset (e.g., the USDA food database).
2. Storing those chunks in a vector database.
3. Writing retrieval logic to query the vectors.

That process is a chore. Instead, I uploaded the USDA files directly to Gemini’s file‑search capability. The model holds the data and can search through its own understanding of it. When a user asks for iron‑rich foods, the AI sifts through the information it already has on file.

Benefits

• No need to manage a separate vector database.
• Setup completed in an afternoon.
• The AI gains better context because it sees the full dataset relationships, not isolated pieces.

Privacy‑First Architecture

When dealing with personal health metrics, privacy isn’t a feature—it’s the foundation. Even as an open‑source project, the architecture enforces data isolation so users can only ever see their own information.

Key decisions

• Users choose which AI provider to use, or supply their own API key.
• This gives users control over where sensitive details are sent, building necessary trust.

Frontend Experience

The interface is built as a Progressive Web App with Inertia, React, and TypeScript. It feels like a native app on your phone, providing a smooth and familiar experience.

Wrapping Up

This process showed that you don’t need the most complex setup to build something both smart and safe. Leveraging newer capabilities in a simpler way can cut out a ton of work, proving that a small team—or even a solo developer—can create a high‑stakes AI tool without getting lost in the infrastructure maze.

Try It Out

• Repository:
• Live demo:

If you found this helpful, consider dropping a ⭐ on the repo!