Atlas Platform¶
Atlas is the unified platform replacing per-product backends across Dapper Collectibles. It consolidates NBA Top Shot, NFL All Day, and Disney Pinnacle onto a shared infrastructure with a common design system, configuration layer, and content management approach.
Why Atlas¶
Before Atlas, each collectibles product ran largely independent backend and frontend stacks. Content configuration lived in Contentful and Retool. Changes required engineering deploys. There was no way to experiment rapidly with page layouts, campaign configurations, or product features without code changes.
Atlas solves three problems:
- Unified platform — one infrastructure serving all three collectibles products, reducing maintenance burden and enabling cross-product features.
- Configuration over code — product and campaign teams can create and modify experiences without engineering deploys.
- AI-native experimentation — the design system is represented as JSON, which means AI agents can generate, modify, and A/B test layouts programmatically.
Architecture Overview¶
graph TD
A[Atlas Design System] --> B[json-render]
B --> C[Atlas Frontend - React]
D[Atlas API - Backend] --> C
E[Campaign Builder UI] --> D
F[AI Campaign Agent] --> E
G[Feature Flags / VPN Gating] --> C
D --> H[Product Data - NBA/NFL/Disney]
Key Components¶
| Component | Description | Repo |
|---|---|---|
| Atlas Frontend | React app consuming JSON layout definitions | atlas-app |
| Atlas API | Backend serving product data and configuration | atlas-api |
| Campaign Builder | WYSIWYG editor for creating campaigns and pages | Built on Atlas staging |
| json-render | Library that renders JSON layout definitions using actual design system components | json-render.dev |
How json-render Works¶
The core innovation is generative UI via JSON. Instead of building pages in code, pages are described as JSON documents that reference components from the Atlas design system. The json-render library takes that JSON and renders it with actual React components.
This means:
- A campaign manager can create a page layout in the Campaign Builder and see it rendered in real-time
- An AI agent can generate page JSON, and it renders identically to hand-built pages
- A/B testing is trivial — serve different JSON to different user segments
- Changes are instant — no build, no deploy, just update the JSON
What's on Atlas vs. Legacy¶
| Feature/Product | Infrastructure | Status |
|---|---|---|
| Quest (challenges) | Atlas | Production — fully live, no issues |
| Top Shot This (TST) | Legacy -> Atlas | Migrating — directed to move immediately as first NBA production feature on Atlas |
| NBA Campaigns | Legacy | Can be created in Atlas tooling to start working on Atlas |
| NFL All Day | Legacy | Pending |
| Disney Pinnacle | Legacy | Pending |
| Campaign Builder | Atlas staging | Staging — 2 PRs pending for v1 (FE: atlas-app#761, BE: atlas-api#632) |
Campaign Builder¶
The Campaign Builder is the primary tool for creating and managing campaigns on Atlas. Built by Jim Wheaton (AI team).
What It Does¶
- WYSIWYG editing of campaign pages using Atlas design system components
- Real-time population — changes are visible as they are made
- VPN-gated access on staging and production (Alex Grieco's WAF/VPN work)
- Generates JSON layout definitions that json-render consumes
Current Status¶
- Demoed successfully to Leon Li, Spencer Bogad, and Jordan Wilberding
- 2 PRs pending for v1 (frontend and backend)
- Working on Atlas staging environment
- Auth0 admin access + VPN required for access
The AI Experimentation Vision¶
The Campaign Builder is the bridge between AI intelligence and production changes. The target architecture:
Heimdall detects signal (e.g., engagement drop in segment)
-> Intelligence layer composes response (new campaign targeting that segment)
-> Campaign Builder generates the page as JSON
-> Feature-flagged deployment to targeted users
-> Heimdall measures result
-> Feedback loop refines the next campaign
This is what "AI-driven autonomous experimentation" means concretely — the ability to compose, deploy, and measure product experiences without engineering deploys for each iteration.
Migration Roadmap¶
| Phase | Scope | Status |
|---|---|---|
| 1 | Quest on Atlas | Complete — production, no issues |
| 2 | TST to Atlas | In progress — CEO directive to move immediately |
| 3 | NBA campaigns on Atlas | Next — campaigns just need to be created in Atlas tooling |
| 4 | NFL and Disney products | Planned |
| 5 | Full platform consolidation | Target state |
Engineering Contact
Atlas migration is jointly owned by Eric Lin (VP Engineering) and Matt Schorr (Executive Producer, Collectibles). The Campaign Builder is owned by Jim Wheaton on the AI team.
Access¶
- Staging: Requires VPN + Auth0 admin credentials
- Production: VPN-gated for internal tooling
This section needs enrichment from the engineering team
Specific API documentation, deployment procedures, and detailed schema information for Atlas should be added by the engineering team. The architecture diagrams above are derived from meeting notes and design documents, not from direct codebase inspection.