AI University: 3-File Pattern to Add Any AI Provider + Offline Fallback Content
What Is AI University
θͺεζ ͺεΌδΌη€Ύ includes "AI University" β a tab-based learning UI for 20+ AI providers (OpenAI, Google, Anthropic, Meta, xAI, Mistral, DeepSeek, etc.):
- Overview / Models / API / News tabs per provider
- Quiz for comprehension check
- Learning scores stored in Supabase + leaderboard
- Share card generation (RenderRepaintBoundary β PNG)
- Consecutive learning streaks
This post covers the 3-file pattern to add any new provider in under 15 minutes.
The 3-File Pattern
Adding a new AI provider requires exactly 3 changes:
1. Supabase Migration (seed SQL)
-- supabase/migrations/20260412000001_seed_reka_ai_university.sql
INSERT INTO ai_university_content (provider, category, title, content, published_at)
VALUES
('reka', 'overview',
'Reka AI β Multimodal-First LLM',
'## What is Reka AI\n\nReka AI is a 2023 AI startup specializing in multimodal models...',
'2026-04-12'),
('reka', 'models',
'Reka Core / Flash / Edge Model Family',
'## Model Overview\n\n### Reka Core\n- Largest model in the family...',
'2026-04-12')
ON CONFLICT (provider, category) DO UPDATE
SET title = EXCLUDED.title,
content = EXCLUDED.content,
published_at = EXCLUDED.published_at;
ON CONFLICT ... DO UPDATE makes the migration idempotent β safe to re-run.
2. Flutter _providerMeta Map
// lib/pages/gemini_university_v2_page.dart
static const Map<String, _ProviderMeta> _providerMeta = {
// ... existing entries
'reka': _ProviderMeta(
displayName: 'Reka AI',
emoji: 'π¦',
color: Color(0xFF7B2FF7),
),
};
This controls how the provider tab renders: name, emoji, accent color. If omitted, the tab still appears (driven by DB), just with default styling.
3. Fallback Content
In case the Edge Function is unavailable (network error, EF not deployed yet):
static const Map<String, String> _fallback = {
// ... existing entries
'reka': '''## Reka AI
Multimodal-first AI startup focused on Reka Core, Flash, and Edge models.
**Key strengths**: Image + text understanding, competitive with GPT-4V.
''',
};
The fallback renders as Markdown. Users get content even offline.
20 Providers Added in One Session
| Category | Providers |
|---|---|
| Big Tech | Google, Microsoft, Meta, Apple |
| AI Startups | OpenAI, Anthropic, xAI (Grok), Mistral, Perplexity, DeepSeek |
| Enterprise | Amazon (Nova/Bedrock), Cohere, IBM (watsonx) |
| Cloud/HW | NVIDIA, Samsung, Baidu (ERNIE) |
| Emerging | Groq, Oracle (OCI), Reka AI |
| Asia | Qwen (Alibaba) |
Each took ~10 minutes: write the SQL seed, add _providerMeta entry, add fallback string.
DB-Driven Tab Generation
The UI reads providers from ai_university_content β no hardcoded tab list:
// Fetch distinct providers from DB
final data = res.data as Map<String, dynamic>?;
final providers = (data?['providers'] as List?)
?.cast<String>() ?? [];
// TabController sized to actual provider count
_tabController = TabController(length: providers.length, vsync: this);
Add a provider to the DB β tab appears automatically. Remove from DB β tab disappears. No Flutter rebuild required.
Auto-Update Architecture
The AI University content stays current via a two-layer update pipeline:
| Layer | Frequency | Method |
|---|---|---|
GitHub Actions ai-university-update.yml
|
Every 2 hours | RSS feeds per provider |
| Claude Schedule | Every 4 hours | NotebookLM Deep Research |
Both upsert to the same ai_university_content table. The later-written record wins β Claude Schedule's richer content overwrites the GHA RSS summary.
Offline Fallback vs Live Data
Future<void> _loadContent(String provider, String category) async {
try {
final res = await _supabase.functions.invoke(
'ai-university-content',
body: {'action': 'get', 'provider': provider, 'category': category},
);
final data = res.data as Map<String, dynamic>?;
setState(() => _content = data?['content'] as String? ?? _fallback[provider] ?? '');
} catch (_) {
setState(() => _content = _fallback[provider] ?? 'Content unavailable');
}
}
The fallback chain: EF response β _fallback[provider] β hardcoded error string. Users always see something.
Try it: θͺεζ ͺεΌδΌη€Ύ
