Omni Architect transforms your PRD into validated Mermaid diagrams and Figma assets. The quality of the output depends heavily on how well your PRD is structured. This guide shows you how to write PRDs that maximize the effectiveness of the automated pipeline.
Omni Architect’s parser (Phase 1) looks for specific patterns in your Markdown PRD to extract semantic meaning. Follow these structural guidelines for best results.
The parser tokenizes your PRD by heading levels (H1, H2, H3):
# Project Name## Feature: User Authentication### User Story...### Acceptance Criteria...## Feature: Product Catalog...
Use H2 (##) for features, H3 (###) for subsections like User Stories, Flows, and Acceptance Criteria. This hierarchy helps the parser correctly classify each section.
The parser recognizes user stories using the standard format pattern:
### User StoryComo **[persona]**, quero **[capability]**, para **[benefit]**.(English version)As a **[persona]**, I want to **[capability]**, so that **[benefit]**.
Example:
### User StoryAs a **shopper**, I want to **complete my purchase in 3 steps or less**,so that I have a **fast and frictionless experience**.
Always bold the key elements (persona, capability, benefit) using **bold** syntax. This helps the NER (Named Entity Recognition) extraction identify important domain concepts.
Flows generate the most important diagrams: flowcharts and sequence diagrams. Use numbered lists or explicit flow sections:
## Checkout Flow### Flow: Complete Purchase1. User views shopping cart2. System checks if user is authenticated - If not authenticated → Redirect to Login/Signup - If authenticated → Continue3. User selects shipping address4. System calculates shipping cost5. User selects payment method6. System processes payment - If approved → Create order - If declined → Show error, return to payment7. System sends confirmation email8. System updates inventory
Explicitly mark decision points (“If/Else”) and error paths (“sad path”). These become conditional branches in flowcharts and improve your completeness score.
Use checkbox lists for acceptance criteria. The parser tracks these for completeness scoring:
### Acceptance Criteria- [ ] User can select saved address or add new one- [ ] Shipping cost calculates in real-time- [ ] Support for PIX, credit card, and bank slip- [ ] Automatic confirmation email sent- [ ] Order appears in user's order history within 5 seconds
Help the parser classify requirements by using clear labels:
## Functional Requirements**FR001:** The system must support guest checkout**FR002:** Users must be able to save multiple addresses## Non-Functional Requirements**Performance:** Checkout flow must complete in < 3 seconds (p95)**Security:** All payment data must be PCI-DSS compliant**Availability:** 99.9% uptime SLA
The parser calculates a completeness score (0.0 - 1.0) based on:
Element
Weight
How to Improve
Features with priorities
20%
Add priority levels (High/Medium/Low)
User stories
20%
Use the standard “As a…I want…so that” format
Entities defined
15%
Include entity tables with attributes
Flows documented
15%
Document step-by-step flows with decision points
Acceptance criteria
15%
Add checkbox lists for each feature
Dependencies mapped
10%
State inter-feature dependencies
Non-functional requirements
5%
Include performance, security, availability requirements
If your completeness score is below 0.6, the parser will emit warnings with specific suggestions. Address these before proceeding to diagram generation.
1. User attempts login2. System validates credentials - If valid → Redirect to dashboard - If invalid → Show error, allow retry3. User views personalized dashboard
Omni Architect supports multiple languages. Set the locale parameter to match your PRD:
pt-BR: Portuguese (Brazil)
en-US: English (United States)
es-ES: Spanish (Spain)
If your PRD is in a different language than your locale setting, the system will auto-detect the language and translate labels in diagrams accordingly.
