← Back to Work

ArcticCodex

Subscription AI Chat Platform: Aurora Model that Learns from User Feedback

About This Project

Status: Live subscription platform at arcticcodex.com

ArcticCodex is now a full-featured subscription AI chat platform featuring Aurora—our proprietary AI model that learns and improves from user feedback.

The Vision

We evolved from a documentation search tool into a subscription AI platform that puts learning at the center. Every conversation with Aurora contributes to making the model smarter.

Users rate responses (1-5 stars), and high-quality conversations automatically feed Aurora's training pipeline. Over time, the model becomes increasingly personalized and accurate for our user base.

What We Built

A modern SaaS platform with subscription tiers (Trial/Basic/Pro/Max), token-based rate limiting, real-time chat, and a learning pipeline that continuously improves Aurora.

  • Token-based pricing with 3 token types (fast, smart, maxpower)
  • Integration with 12+ AI models (GPT-4o, Claude, DeepSeek, Gemini, etc.)
  • Aurora model that learns from user feedback (rating ≥4 stars)
  • Automatic training data collection and quality scoring
  • Full Supabase auth with subscription enforcement
  • Analytics dashboard for learning metrics

Tech Stack

Frontend

  • • Next.js 16 (App Router)
  • • TypeScript
  • • Tailwind CSS (black/green theme)
  • • Real-time chat UI

Backend

  • • Supabase (PostgreSQL + Auth + RLS)
  • • Stripe subscription management
  • • Token bucket rate limiting
  • • Learning pipeline (training data collection)

AI/Models

  • • Aurora (proprietary learning model)
  • • GPT-4o, Claude 3.5, DeepSeek-v3
  • • Google Gemini, Grok 3, and others
  • • Feedback-driven model training

Deploy & Ops

  • • Vercel (production)
  • • GitHub Actions CI/CD
  • • Stripe webhooks
  • • Analytics & metrics

Key Features

  • ✅ 4-tier subscription model (Trial/Basic/Pro/Max)
  • ✅ Token-based usage tracking (fast/smart/maxpower)
  • ✅ Real-time chat with multiple AI models
  • ✅ Automatic feedback collection (1-5 star ratings)
  • ✅ Training data pipeline for Aurora
  • ✅ Learning metrics and quality scoring
  • ✅ Full Supabase authentication
  • ✅ Stripe integration for billing

How Aurora Learns

Every conversation is tagged as Aurora. When users rate responses (≥4 stars), those high-quality exchanges automatically enter the training dataset. This feedback loop makes Aurora progressively smarter.

100%
Conversations tracked
Auto
Training data collection
Monthly
Model improvements

Results & Metrics

  • • Users can choose from 12+ AI models based on tier
  • • High-quality feedback drives Aurora's continuous learning
  • • Stratified training data sampling ensures model diversity
  • • Clear ROI: users see model improvement month-over-month
  • • Transparent pricing with no surprise costs

What This Demonstrates

  • • Full-stack SaaS with subscription management and Stripe integration
  • • Complex database schema with RLS policies for security
  • • Real-time chat frontend with React state management
  • • Multi-model AI integration (OpenAI, Anthropic, etc.)
  • • Feedback loops and continuous learning architecture
  • • Production-grade token bucket rate limiting
  • • Complete CI/CD and Vercel deployment