Full Lifecycle Workflow

Step 1.1: Define Document Scope

Before generating content, clearly define your document requirements:

Step 1.2: Select Artifact Template

1. Navigate to the Authoring page
2. Click Artifact Template dropdown
3. Review available templates:
   • Technical Specification: Best for API docs, architecture decisions
   • Product Requirements: ✅ Best for PRDs, user stories, feature specs
   • Research Report: Best for data analysis, academic content
   • Marketing Copy: Best for campaigns, landing pages
4. Select Product Requirements template
5. Preview the template structure (Overview, User Stories, Technical Specs, Acceptance Criteria)

Step 1.3: Choose Execution Mode

Select the execution mode based on document complexity:

Decision: Use Mixture of Agents strategy for high-quality PRD with draft → critique → polished composition workflow.

Step 1.4: Configure Model Selection

1. Select execution mode: Strategy
2. Choose chain strategy: Mixture of Agents
3. Assign models to roles:
   • Drafter: Google Fast and Effective (SKU Level 4) - for rapid initial draft
   • Critic: Claude little song (SKU Level 9) - for thorough analysis
   • Composer: Openai leader and effective (SKU Level 9) - for polished final output
4. Set template tone: Balanced (temperature: 0.7)
5. Set max output tokens: 4000 (for comprehensive content)

Step 2.1: Write Comprehensive Prompt

Craft a detailed prompt that provides all necessary context:

Create a comprehensive Product Requirements Document for the following feature:

Feature: In-App Voice Messaging with Transcription

Context:
- Existing chat app with 500K MAU
- Users requesting voice messages for hands-free communication
- Must support iOS and Android
- Privacy-critical: all transcriptions must be on-device

Target Audience:
- Engineering Team: Need technical architecture, API specs, data models
- Product Stakeholders: Need business justification, user stories, success metrics
- QA Team: Need acceptance criteria, test scenarios, edge cases

Required Sections:
1. Executive Summary (business value, success metrics)
2. User Stories (as a [user], I want [feature], so that [benefit])
3. Technical Architecture (client-side, backend, ML model)
4. API Specifications (endpoints, request/response formats, error codes)
5. UI/UX Flow (wireframe descriptions, user journey)
6. Data Model (database schema, message storage, metadata)
7. Security & Privacy (encryption, on-device transcription, GDPR compliance)
8. Acceptance Criteria (functional, performance, security requirements)
9. Out of Scope (clarify what this release does NOT include)
10. Timeline & Dependencies (phases, external dependencies)

Technical Constraints:
- Must work offline (queue messages, sync when online)
- Max voice message length: 3 minutes
- Transcription accuracy target: 95% WER (Word Error Rate)
- Must comply with GDPR, CCPA data privacy regulations

Please generate a detailed, professional PRD suitable for engineering implementation.

Step 2.2: Execute Generation

1. Review all settings (models, template, tone, max tokens)
2. Click Generate button
3. Monitor progress through 3 stages:
   • Stage 1 (Drafter): Initial rough draft (~90 seconds)
   • Stage 2 (Critic): Analysis and improvement suggestions (~60 seconds)
   • Stage 3 (Composer): Polished final output (~120 seconds)
4. Total generation time: ~4-5 minutes

Step 3.1: Review All Three Outputs

1. The AI Generation Pane displays three tabs:
   • Drafter Output: Review for content completeness
   • Critic Output: Read improvement suggestions
   • Composer Output: Final polished PRD (primary deliverable)
2. Evaluate the Composer output against requirements:
   • ✓ All required sections present?
   • ✓ Technical details accurate and complete?
   • ✓ Tone appropriate for audience?
   • ✓ Acceptance criteria specific and testable?
3. Compare Drafter vs Composer to see improvement quality
4. Note any sections that need significant manual refinement

Step 3.2: Save Baseline Version

1. Click Save Version in the toolbar
2. Enter version name: "v0.1 - Initial Generation (Mixture of Agents)"
3. Add version notes:

   Initial PRD generation using Mixture of Agents strategy
   Drafter: Google Fast and Effective
   Critic: Claude little song
   Composer: Openai leader and effective
   
   Status: First draft, needs review for technical accuracy
   Next Steps: Validate API specs with backend team

4. Click Save
5. Verify success message: "Version saved successfully"

Step 4.1: Identify Sections Needing Work

Review the Composer output and create a refinement plan:

Step 4.2: Regenerate Weak Sections

For the API Specifications section that needs significant improvement:

1. Hover over the "API Specifications" section header
2. Click the Regenerate icon (circular arrow)
3. In the regeneration prompt dialog, provide specific guidance:

   Regenerate API Specifications with:
   - Comprehensive error codes (400, 401, 403, 413, 429, 500, 503)
   - Rate limiting details (max 100 requests/minute)
   - Sample request/response JSON for each endpoint
   - Authentication headers (Bearer token format)
   - Pagination for list endpoints (cursor-based)
   - File upload specs for voice messages (max 10MB)

4. Select model: GPT-4 Turbo (same as Composer for consistency)
5. Click Regenerate
6. Review new output and accept if improved

Step 4.3: Manual Editing for Domain Expertise

For sections needing domain-specific refinement (Technical Architecture, Security):

1. Click on the "Technical Architecture" section to activate Tiptap editor
2. Add specific ML model recommendation:

   Transcription Engine:
   - iOS: Apple Speech Framework (on-device, 60+ languages)
   - Android: ML Kit Speech-to-Text (on-device, 50+ languages)
   - Fallback: OpenAI Whisper (cloud-based, 99 languages)
   
   Model Selection Strategy:
   - Check device support for on-device transcription
   - If supported: Use native framework (privacy-first)
   - If not supported: Prompt user for cloud transcription consent

3. Use formatting toolbar: headings, bullet lists, code blocks
4. Click outside section to deactivate editor and save changes

Step 4.4: Enhance with Visual Elements

1. For the UI/UX Flow section, add wireframe placeholders:
   • Insert image placeholders for key screens
   • Add captions: "Figure 1: Voice message recording screen"
   • Link to external design tool (Figma, Sketch)
2. For Data Model section, format tables:
   • Use table formatting for database schema
   • Include columns: Field Name, Type, Constraints, Description

Step 4.5: Save Refined Version

1. After completing all refinements, click Save Version
2. Version name: "v0.2 - Post-Refinement"
3. Version notes:

   Refinements applied:
   - Regenerated API Specifications with comprehensive error codes
   - Added specific ML model recommendations (Apple Speech, ML Kit, Whisper)
   - Enhanced Security section with GDPR data retention policies
   - Added 3 new user stories for edge cases (offline, interruptions)
   
   Status: Technical review ready
   Next Steps: Share with engineering lead for architecture validation

4. Click Save

Step 5.1: Export Draft for Review

1. Click Export in the toolbar
2. Select format based on reviewer needs:
   • PDF: For executive stakeholders (read-only, formatted)
   • DOCX: For engineering team (editable, collaborative comments)
   • HTML: For web publishing or Confluence import
3. Enable "Include Model Attribution" for transparency
4. Add watermark: "DRAFT v0.2 - For Review Only - [Date]"
5. Export and share via email/Slack/document management system

Step 5.2: Collect Feedback

Create a structured feedback collection process:

Step 5.3: Incorporate Feedback

1. Return to xeditai and load version v0.2 (if not already active)
2. Address each feedback item systematically:
   • API rate limits: Edit API Specifications section, update to 300 req/min
   • Redis caching: Edit Technical Architecture, add caching layer diagram
   • Timeline: Edit Timeline section, extend to 8 weeks, add buffer
   • Success metrics: Regenerate Executive Summary with specific KPIs
   • Competitive analysis: Add new section with comparison table
   • CCPA compliance: Edit Security section, add deletion flow
3. For complex additions (competitive analysis), use Single mode generation:
   • Select section to insert after
   • Use slash command / to insert new section
   • Prompt: "Create competitive analysis comparing our voice messaging feature to WhatsApp, Telegram, and Signal"

Step 5.4: Save Post-Review Version

1. Click Save Version
2. Version name: "v0.3 - Post-Stakeholder Review"
3. Version notes:

   Incorporated feedback from engineering, product, and legal reviews:
   
   Engineering:
   - Updated API rate limits (100 → 300 req/min)
   - Added Redis caching layer to architecture
   - Extended timeline to 8 weeks with buffer
   
   Product:
   - Added success metrics (20% DAU increase, 50K voice messages/day)
   - Added competitive analysis section (WhatsApp, Telegram, Signal)
   
   Legal/Compliance:
   - Added CCPA-specific data deletion flow
   - Clarified data retention: 30 days default, user-configurable up to 1 year
   - Added requirement for consent flow screenshot
   
   Status: Ready for final approval
   Next Steps: Schedule approval meeting with stakeholders

4. Click Save

Step 6.1: Final Quality Check

Perform comprehensive quality assurance before marking as final:

Step 6.2: Polish Formatting

1. Review entire document for visual consistency:
   • Heading hierarchy: H1 for document title, H2 for main sections, H3 for subsections
   • Bullet list formatting: consistent style (bullets vs numbers)
   • Code blocks: use monospace font for all technical snippets
   • Tables: ensure consistent column widths and alignment
2. Add document metadata:
   • Document title, version, date
   • Authors and contributors
   • Approval signatures (if required)
   • Revision history table
3. Enable Block Mode for advanced layout (if needed):
   • Adjust section widths for optimal readability
   • Center important diagrams or callouts
   • Use side-by-side blocks for comparison tables

Step 6.3: Save Final Version

1. Click Save Version
2. Version name: "v1.0 - FINAL APPROVED"
3. Version notes:

   Final approved version of Voice Messaging PRD
   
   Approved by:
   - Sarah Chen (Engineering Lead) - [Date]
   - Mike Rodriguez (Product Manager) - [Date]
   - Jennifer Wu (Legal/Compliance) - [Date]
   
   Changes from v0.3:
   - Removed DRAFT watermark
   - Added approval signatures
   - Final formatting polish
   - Added revision history table
   
   Status: FINAL - Ready for Implementation
   Distribution: Engineering team, product stakeholders, QA team

4. Click Save

Step 7.1: Export Final Deliverables

1. Create multiple export formats for different audiences:
   • PDF (with model attribution off): For official distribution, archival
   • DOCX: For team collaboration, future edits
   • HTML: For internal wiki/Confluence publishing
2. Name files with version and date:
   • Voice_Messaging_PRD_v1.0_2026-01-25.pdf
   • Voice_Messaging_PRD_v1.0_2026-01-25.docx
   • Voice_Messaging_PRD_v1.0_2026-01-25.html
3. Verify each export:
   • PDF: Check formatting, images render correctly, page breaks logical
   • DOCX: Verify tables, ensure headers/footers correct
   • HTML: Test in browser, verify links functional

Step 7.2: Distribute to Stakeholders

1. Upload to document management system:
   • Google Drive / SharePoint: Set permissions (view-only for PDF, edit for DOCX)
   • Confluence: Import HTML version to product wiki
   • Jira: Attach PDF to epic ticket
2. Send distribution email:

   Subject: [FINAL] Voice Messaging PRD v1.0 - Approved for Implementation
   
   Team,
   
   The Voice Messaging with Transcription PRD is now finalized and approved.
   
   📄 Document: [Link to PDF]
   ✏️ Editable Version: [Link to DOCX]
   🌐 Wiki: [Link to Confluence page]
   
   Key Highlights:
   - Timeline: 8 weeks (3 sprints)
   - Success Metrics: 20% DAU increase, 50K messages/day
   - Privacy-First: On-device transcription (iOS/Android)
   - Compliance: GDPR & CCPA compliant
   
   Next Steps:
   1. Engineering kickoff meeting: [Date/Time]
   2. Sprint planning: [Date/Time]
   3. Design handoff: [Date/Time]
   
   Questions? Reply to this thread or ping me on Slack.
   
   Thanks,
   [Your Name]

Step 7.3: Archive in xeditai

1. In xeditai, add a final archive note to the document
2. Tag the document for easy retrieval:
   • Tags: "PRD", "Voice Messaging", "v1.0", "2026-Q1", "Approved"
3. Set document status: Published or Archived (if available)
4. Ensure all versions (v0.1, v0.2, v0.3, v1.0) are preserved for audit trail

Step 8.1: Post-Implementation Updates

As the feature is implemented, the PRD may need updates:

1. Load version v1.0 in xeditai
2. Make necessary updates:
   • API endpoints changed during implementation
   • New edge cases discovered in development
   • Performance metrics updated based on testing
   • Security vulnerabilities addressed
3. Save as v1.1 with clear notes about what changed
4. Re-export and update distributed versions
5. Notify stakeholders of updates via email/Slack

Step 8.2: Periodic Reviews

Schedule regular document reviews to keep content current:

Step 8.3: Reuse as Template

1. If this PRD was successful, save it as a template for future PRDs:
   • Remove feature-specific content
   • Keep section structure, formatting, approval flow
   • Add placeholder guidance text for each section
   • Document lessons learned in template notes
2. Share template with product team for consistency
3. Iterate on template based on feedback from multiple uses