01 / The question
What if prompt work felt more like mapping than guessing?
The core idea was to give each experiment a place to live. That meant making the relationship between an input, a model, and an output easy to scan at a glance.
Back to projects
01AI research playground
Prompt Atlas
A visual workspace for comparing prompts, models, and the unexpected paths between them.
Role
Product design + full-stack prototype
Timeline
6 weeks
Status
Independent experiment
Stack
Next.js / AI / Product design
The challenge
AI tools are powerful, but the path from a first prompt to a useful result can feel like a black box. I wanted to make the process visible enough to learn from.
The approach
I designed a focused workspace where prompts, model outputs, and useful discoveries could sit beside each other instead of disappearing into a chat history.
The outcome
A clear prototype for treating AI exploration as an iterative design practice rather than a sequence of lucky guesses.
Inside the work
02 / The system
A workbench for side-by-side thinking.
The interface uses a compact visual language for model settings, prompt versions, and notes. It keeps the technical details close without making the whole experience feel like a developer console.
Key beats
01Prompt versioning
02Model comparison
03Experiment notes
Visual notes
A closer look.
Next case study