PromptZero

Inspiration

AI is powerful, but communicating with it is still surprisingly inefficient. People repeat themselves, use vague wording, forget important constraints, and attach documents in formats that are difficult to reuse. Creative users face another problem: they may know exactly how a video should look without knowing terms such as dolly-in, low-angle shot, or *shallow depth of field

What it does

PromptZero helps people turn messy, wordy prompts into clearer instructions before sending them to AI.

How we built it

Built with Jac nodes, edges, and walkers, it extracts goals and constraints into an inspectable intent graph, flags ambiguity, and suggests conservative rewrites. Its Meaning Lock checks that detected critical requirements remain intact, while a Context Diff shows what was preserved, compressed, or removed.

Challenges we ran into

The hardest part was shortening prompts without changing their meaning. ## Accomplishments that we're proud of The Safari extension let me review, apply, copy, or undo optimizations without automatically sending messages. I also added PDF-to-Markdown conversion, camera vocabulary assistance

What we learned

Prompt optimization is closer to compilation than summarization.

What's next for PromptZero

Build a larger evaluation dataset.

Built With

• and
• chrome
• css
• extension
• fastapi
• html
• jac
• javascript
• python