Promptheus - Teams were relying on AI without knowing which prompts actually worked. Promptheus gave them one place to test, improve, share, and reuse what gets better results.

6Scoring dimensions

2Model providers

4Org roles

1Shared workspace

The problem

Prompt quality was mostly guesswork.

Teams were writing more prompts, reusing more prompts, and depending more heavily on LLMs, but the workflow around those prompts was still surprisingly primitive. A prompt usually lived in a document, a chat history, or someone’s notes. Quality was subjective. Good prompts were hard to reuse. Collaboration was messy. There was no consistent way to score what worked, explain why it worked, or turn one person’s prompt into something a whole team could build on. Promptheus started as an internal hackathon attempt to make that workflow more structured. What began as a week-long build became a full product: a multi-model workspace where prompts can be written, tested, scored, improved, saved, templated, shared, and reused across teams. The goal wasn’t to create another prompt library. It was to build the workflow around prompt quality.

How it runs · 06 steps

Write the PromptThe user works inside a chat-style editor and chooses the model they want to test against.

trigger

Run Against Multiple ModelsThe same workspace can dispatch requests to OpenAI or Anthropic without changing the frontend workflow.

Score Across Six DimensionsA scoring engine evaluates clarity, specificity, structure, completeness, tone, and context.

Surface ImprovementsThe system returns ranked suggestions and contradiction flags alongside the score.

process

Save and OrganizePrompts can be stored in personal or organization libraries, grouped into collections, and tagged automatically.

storage

Reuse or PublishSaved prompts can become reusable templates or be published to the community marketplace for discovery, rating, comments, and forks.

output

Engineering detail · 04

One workspace, multiple modelsPromptheus treats model providers as interchangeable at the product layer. A single frontend workflow can stream responses from OpenAI or Anthropic while keeping the user experience consistent. The architecture keeps provider-specific logic behind the API boundary rather than coupling the product to one model vendor.

Scoring that explains itselfEach prompt is scored across six dimensions — clarity, specificity, structure, completeness, tone, and context. The system does more than return a number. It also detects contradictions and produces ranked, actionable suggestions so the user can understand what needs to improve and why.

Prompts become reusable systemsA template layer turns prompts into reusable assets instead of static text. {{placeholder}} syntax supports variable extraction, validation, deduplication, and live substitution previews, making one good prompt reusable across many inputs and workflows.

Built for teams, not just individualsPromptheus includes organization workspaces, shared libraries, member roles, folder-level permissions, invites, and usage visibility. The marketplace adds another layer of reuse — prompts can be published, discovered, rated, commented on, saved, and forked into new templates.

Result

Promptheus started as a one-week internal hackathon project. Instead of stopping at the prototype, we kept pushing it into a real product: multi-model streaming, prompt scoring, reusable templates, organization access control, shared libraries, marketplace mechanics, and analytics. The project became less interesting because of any single AI feature and more interesting because of everything required around the AI to make the product coherent, collaborative, and production-ready. The model call was the easy part. The product around it was the real build.

NextPulse →We automated work inside legacy software that has no API, from booking appointments to monitoring records, without anyone needing to operate the software manually.

Bring me a problem like this →