PRD-004: Prompt Engineer Training (PET) App
1. Product Vision
Prompt Engineer Training (PET) is a dual-pane, gamified learning platform that transforms users from novices to expert prompt engineers. It provides immediate, actionable feedback on prompt intention, structure, and effectiveness, specifically targeting both conversational and complex agentic (multi-step) contexts.
Core Differentiators:
- Prompt Judge: Evaluating the prompt itself, not just the output.
- Agentic Focus: Teaching orchestration, tool usage, and constraints.
- Gamified Learning: Skill-based progression (Duolingo style).
- Enterprise Grade: Privacy-first, on-premise capable, with domain-specific rule enforcement.
2. Target Audience
- Individuals: Self-improvers, students seeking clear, consistent AI interaction skills.
- Enterprise ICs: Knowledge workers needing to follow org policies and improve efficiency.
- L&D Organizations: Teams needing to scale prompt literacy and track ROI.
3. User Experience
3.1 Dual-Pane Interface
- Left Pane: Prompt input & raw AI response.
- Right Pane: “Prompt Judge” feedback panel showing:
- Inferred Intent
- Clarity/Specificity Score
- Agentic Effectiveness
- Mental model gaps
- Concrete improvement suggestions
3.2 Key Workflows
- Submit Prompt: User enters prompt.
- Get Judged: System generates Response (Left) and Evaluation (Right).
- Iterate: User clicks “Apply Improvements” or manually edits.
- Learn: System logs progress, awards XP, and triggers micro-lessons if specific weaknesses are detected (e.g., “Missing Constraints”).
4. Feature Requirements
4.1 Core Features (MVP)
| ID | Feature | Description |
|—|—|—|
| FEAT-001 | Modular Prompt Judge | Pipeline to evaluate intent, structure, constraints, and agentic viability. |
| FEAT-002 | Feedback UI | Structured rubric display (score 1-5, explanation, fix). |
| FEAT-003 | Auto-Improve | Button to regenerate the prompt with suggested fixes applied. |
| FEAT-004 | Lesson Library | Curated “micro-lessons” triggered by specific failure flags. |
| FEAT-005 | Gamification | XP, streaks, and skill badges (e.g., “Constraint Master”). |
| FEAT-006 | Multilingual Support | Judge auto-detects language and provides feedback in the user’s native tongue. |
| FEAT-007 | Desktop App | Native shell (Tauri) for specialized prompt engineering workflow. |
4.2 Enterprise Features (Post-MVP)
| ID | Feature | Description |
|—|—|—|
| ENT-001 | Custom Rubrics & Rules | Org-specific “Best Practice” publishing (versioned). |
| ENT-002 | Team Analytics | Leaderboards and skill heatmaps for managers. |
| ENT-003 | SSO & RBAC | Integration with enterprise auth + role-based access. |
| ENT-004 | On-Prem Deployment | Containerized install with field-level encryption. |
| ENT-005 | LMS Export | SCORM/xAPI packages for learning data portability. |
4.3 Gamification System (FEAT-005)
XP Calculation
| Action |
XP Earned |
Multiplier |
| Submit prompt |
5 |
1x |
| Improve prompt (score +10) |
15 |
1x |
| Complete lesson |
25 |
1x |
| First prompt of day |
10 |
Streak bonus |
| Perfect score (100) |
50 |
1x |
Streak Bonus Formula: base_xp × (1 + streak_days × 0.1) (max 2x)
Skill Badges
| Badge |
Criteria |
Tier |
| Intent Master |
10 prompts with intent score ≥ 4 |
Bronze |
| Constraint Expert |
25 prompts with all constraints met |
Silver |
| Agentic Architect |
50 agentic prompts with score ≥ 80 |
Gold |
| Streak Champion |
30-day continuous streak |
Platinum |
Leaderboards
- Scope: Org-level (enterprise), Global (opt-in)
- Metrics: Weekly XP, Improvement Rate, Badge Count
- Privacy: Users can opt out at any time
4.4 Lesson Trigger Rules (FEAT-004)
| Flag |
Lesson Triggered |
Description |
#missing_constraints |
“Defining Output Constraints” |
Teaches format, length, and safety constraints |
#agentic_ambiguity |
“Agentic Prompt Clarity” |
Tool parameters, error handling |
#vague_intent |
“Sharpening Your Intent” |
Single vs. multi-goal prompts |
#no_examples |
“Power of Few-Shot” |
Adding examples to prompts |
5. Non-Functional Requirements
- Privacy: End-to-end (field-level) encryption for stored prompts.
- Latency: Judge feedback < 5s (streaming).
- Scalability: Nx Modular Monolith architecture to support multiple clients.
- Modularity: Hexagonal architecture (Ports & Adapters) for core domain logic.
6. Success Metrics
- Prompt Improvement Rate: % of sessions where the second prompt scores higher than the first.
- Retention: D30 retention for gamified users.
- Global Reach: % of prompts processed in non-English languages.
- Enterprise Adoption: Number of active seats/teams.
7. Roadmap Strategy
- Phase 1 (MVP): Web App (Next.js), Multilingual Judge, Core Gamification.
- Phase 2: Desktop App (Tauri) & Enterprise Pilot (SCORM, Custom Rubrics).
- Phase 3: Mobile App & Advanced Agentic Scenarios.
8. Non-Goals (v1)
- Direct Slack/GitHub Integrations: We focus on training, not workflow automation.
- Fully Generative Curriculum: All lessons must be strictly human-templated.