SCHWEP
Precision Design Discovery Engine
SCHWEP is a visual design discovery engine that helps me explore and define my aesthetic without writing prompts. Instead of describing what I want, I interact with designs directly, and SCHWEP learns my style through four fast discovery rounds.
By the end, I get a complete design blueprint including layout rules, typography, component tokens, and live mockups that I can export into Figma using my custom plugin or a script that generates SVGs that can be dropped directly into design tools. The codebase is structured so it can be opened in an IDE and run locally after installing dependencies.
Inspiration
Most AI design tools rely heavily on text prompts. I found that frustrating because I know what I like visually but often struggle to describe it.
I built SCHWEP to flip that problem: instead of explaining my taste, I discover it visually. The system learns my style through interaction, letting me focus on what feels right rather than what I can articulate.
What It Does
SCHWEP guides me through four discovery rounds.
Frame
I select structural foundations such as bento grids, Swiss grids, or hero canvas layouts to establish core organization.
Shape
I define geometric identity by adjusting border radii and edge styles to capture the silhouette that matches my taste.
Tone
I explore typography pairings that express different voices, from bold technical styles to editorial elegance.
Finish
I apply atmospheric polish such as glow effects, grain textures, or glass-style depth.
As I make choices, SCHWEP updates my aesthetic vector profile and surfaces designs that are mathematically closest to my evolving style.
How I Built It
SCHWEP runs as a lightweight web application using vanilla JavaScript and modern CSS for fast responsive interaction.
Vector Similarity Engine
Every design card is encoded as a multi-dimensional vector.
When I select or like a design, my profile vector updates and cosine similarity is used to rank catalog designs so that the most aligned designs appear first.
AI Variant Generation
I used Gemini 1.5 Flash to generate new design variations based on my learned aesthetic profile. This allowed me to explore UI concepts that were not in the catalog but still matched my style.
Custom Rendering Tool
I built a custom script that generates SVG layouts from SCHWEP blueprint outputs.
These SVGs can be exported and dropped directly into Figma or other design environments.
The codebase is organized so it can be opened in an IDE and run locally after installing dependencies.
Antigravity
Since I was working alone, Antigravity was extremely helpful.
Running several AI agents in parallel made development feel collaborative even though I was the only developer.
Different agents explored algorithm design, interaction logic, and variant generation ideas simultaneously.
This parallel workflow allowed me to iterate faster and experiment more freely than working sequentially.
Deployment
The project is prepared for hosting and testing environments.
Challenges I Faced
Modeling subjective design taste numerically was challenging. I experimented with balancing aesthetic nuance without making the vector model unnecessarily complex.
Maintaining real-time responsiveness while updating rankings and generating variants was also difficult.
Integrating the rendering pipeline so blueprint outputs could be used directly in design tools required careful handling of layout generation.
Accomplishments I’m Proud Of
I built a system that allows design discovery without prompt writing.
I created a pipeline that connects discovery directly to implementation through SVG export.
Using Antigravity, I was able to develop the project as if I had a team working alongside me.
What I Learned
Visual preference can be represented using mathematical feature spaces.
Preference-based discovery can be more intuitive than prompt-based workflows.
Working with multiple AI agents in parallel significantly accelerated my development process.
What’s Next for SCHWEP
I want to expand the design catalog and refine the aesthetic vector model.
I also plan to improve the rendering pipeline so blueprint outputs can generate more complex and fully editable components.
Long term, SCHWEP could become a collaborative design discovery platform for teams.
Log in or sign up for Devpost to join the conversation.