Architecture Builder guide
Understand how system diagrams, data models, and tech stack recommendations are generated — and how to get the most from them.
What the Architecture Builder generates
The Architecture stage produces a comprehensive system design document with 6 key sections:
System Architecture
A high-level diagram showing how your core modules connect — frontend, backend, database, cache, auth, external APIs, and queue layers.
Data Model
Primary database entities, key fields, relationships, and index recommendations for your system.
Tech Stack Recommendation
Frontend, backend, database, authentication, hosting, and CI/CD recommendations based on your wizard preferences.
Module Breakdown
Each core module with its purpose, key components, and where it fits in the overall system.
API Layer Overview
Key REST API endpoints with method, path, and purpose. See Technical Design for full API spec.
Infrastructure
Hosting recommendations (Vercel, Railway, AWS), environment separation, and deployment strategy.
Dependencies
For best results, run Research before Architecture. Nova uses your market and competitor context to make smarter technical decisions — for example, if competitors use real-time features, Nova will suggest WebSockets in the architecture.
If you skip Research, the Architecture Generator still produces excellent output based solely on your wizard context. Research-informed architecture tends to be more targeted.
Customising the tech stack
If the recommended stack doesn't match your preferences, return to the wizard and update the Technical Preferences step, then click Regenerate Architecture. Your new preferences are applied immediately.
- Switching from PostgreSQL → MongoDB
- Switching from Node.js → Python/FastAPI
- Switching from Vercel → AWS or Railway
- Adding Redis for caching or Kafka for event streaming
Using architecture output with AI coding tools
The architecture document is one of the key inputs to your Prompt Pack. The Master Operating Prompt includes a condensed version so your AI coding tool understands the full system context before generating any code.
You can also copy individual sections (like the Data Model) and paste them directly into Cursor or Claude to give focused context for database-layer generation.