50K+
Builders
2M+
Models Trained
Up to 10×
Training Speedup
About the Project
Gaia (Flare Studio) is an agentic AutoML platform built as a two-portal Next.js TypeScript application, a builder-facing ML studio and an admin portal, backed by a FastAPI Python backend, Celery task queue for async model training jobs, Redis as the task broker, and PostgreSQL as the primary datastore. The entire stack is Dockerised and deployed on Hostinger VPS. Google OAuth handles user authentication; Razorpay powers subscription billing; Nodemailer delivers training job completion emails and model deployment notifications. Google Analytics and Google Search Console track the platform's inbound builder acquisition.
How It Works
- 1
The two-portal monorepo runs on Hostinger VPS, Docker Compose orchestrates the Next.js frontend, FastAPI backend, Celery worker fleet, Redis broker, and PostgreSQL database as a single deployable stack with Nginx as the reverse proxy.
- 2
Google OAuth handles user authentication across both portals; PostgreSQL stores user accounts, project definitions, dataset metadata, and training run history under a multi-tenant schema.
- 3
Dataset uploads trigger a Celery task via Redis; the FastAPI orchestrator runs automated profiling (class distribution, missing value analysis, label noise detection) before queuing the AutoML architecture search loop.
- 4
The Celery worker fleet executes Bayesian-optimised architecture search in parallel, runs transfer learning from pre-trained checkpoints, and exports the best model as ONNX, H5, or PKL, notifying the user via Nodemailer on completion.
- 5
Razorpay handles subscription billing with server-side webhook handlers updating entitlement state in PostgreSQL; Google Analytics tracks dataset upload-to-deployment conversion rates across builder cohorts.
Tech Stack
Want to build something like this?
We'd love to hear about your project. Let's talk about what you're building.