.png)
Overview
Promptheus is a full SaaS for writing, optimizing, and managing AI prompts. The workspace supports GPT-4o and Claude 3.5 Sonnet with real-time streaming — users iterate on prompts in a chat-style interface, then save to a personal library organized by collection and AI-generated tags. A scoring engine rates every prompt across six dimensions: clarity, specificity, structure, completeness, tone, and context, then surfaces actionable improvement suggestions. A template system supports {{placeholder}} syntax with AI-powered variable detection. The community marketplace lets anyone publish, discover, rate, and fork prompts by category. Organizations get shared libraries, role-based access (Owner / Admin / Member / Viewer), folder-level permissions, and usage analytics.
How it works
User writes in the multi-model workspace and selects GPT-4o or Claude 3.5 Sonnet per request.
SSE stream from the model API renders tokens in real-time in the chat interface.
Scoring engine rates the prompt on clarity, specificity, structure, completeness, tone, and context.
Ranked improvements and contradiction flags returned alongside the score report.
Prompt saved to personal library; AI generates tags and assigns it to a collection.
Optionally published — discoverable by category, rateable, commentable, and forkable as a template.
Results
What was built
Multi-model workspace — GPT-4o and Claude 3.5 Sonnet with streaming SSE; frontend is model-agnostic, a single API route dispatches to either provider.
6-dimension scoring engine — rates prompts on clarity, specificity, structure, completeness, tone, and context. Runs contradiction detection and returns ranked improvement suggestions.
Template system — {{placeholder}} syntax with AI-powered variable detection; deduplication, snake_case validation, and live substitution preview.
Community marketplace — publish, browse by category, rate (1–5 stars), comment, save to personal library, and fork into templates. Aggregate metrics prevent duplicate ratings per user.
Organization tier — shared libraries with folder-level permissions, role-based access control, member invites via code, and per-org usage analytics.
Freemium model — free up to 50 org prompts and 5 members; Pro and Enterprise tiers unlock higher limits and advanced features.
Role
Co-built during a week-long internal hackathon, then took it through productionization — shipping the scoring engine, marketplace social layer (ratings, comments, saves, favorites), organization membership model with RBAC, template system with variable detection, and analytics dashboard.
Stack