AI Lab
Where a product designer learns to ship code. Each project started as a question—and ended with a working prototype.
This is how it starts. A question, a prompt, and curiosity.
My Approach
I don't use AI to skip the design process—I use it to iterate faster on more ambitious ideas.
Intent First
Define the problem clearly before prompting. The better I articulate what I want, the better the output.
Fast Loops
Prototype in hours, not weeks. Speed enables risk-taking—ambitious ideas become low-stakes experiments.
Ship & Learn
Real users teach more than assumptions. Deploy early, gather feedback, iterate with purpose.
EqualTales
Can AI create meaningful children's content that challenges stereotypes?
The Challenge
A hackathon challenge with a mission: create something that helps close the gender gap in STEM.
What I Tried That Didn't Work
- Generating unique characters per story—faces were inconsistent across pages
- Using photorealistic image style—uncanny valley effect
- Generic fictional role models—felt hollow without authenticity
What Actually Worked
- Real historical women (Marie Curie, Mae Jemison) for instant authenticity
- Watercolor illustration style for cohesive, warm visuals
- Child's name woven into narrative for emotional connection
Key Trade-off
Chose authenticity over flexibility—real historical figures limit story variety but create deeper emotional impact.
HafsaUsmani.com
Can a designer build a custom portfolio without writing code from scratch?
The Challenge
Create a portfolio that showcases both design sensibility and technical capability—standing out from template-based sites.
What I Tried That Didn't Work
- Using page builders—too generic, no custom interactions
- Hand-coding everything—too slow, constant context switching
- Off-the-shelf templates—couldn't showcase actual skills
What Actually Worked
- Prompt-driven development with Claude Code for complex logic
- Iterative conversation to refine animations and effects
- Custom CSS over framework constraints for full control
Key Trade-off
Chose custom build over speed—took longer but resulted in a portfolio that demonstrates the very skills it showcases.
Other Experiments
Quick builds that validated ideas and taught lessons.
Interview Sage
View LiveCan AI provide meaningful interview practice and feedback?
Speed dramatically lowers the barrier to shipping useful products. Ideas don't have to stay ideas—they can become working tools.
HeadshotAI
View LiveCan no-code AI tools create professional-grade image applications?
Visual builders force clearer thinking about user flow. Constraints aren't limitations—they're design decisions made for you.
AI Toolbox
Different tools for different jobs. Here's what I've learned about each.
Claude Code
Pair programming with AI that understands context across files
Cursor
Best for iterative refinement when you know what you want
Replit Agent
Lowest friction from idea to deployed app
Lovable
Perfect for validating ideas without writing code
DALL-E
Style consistency requires careful prompting and iteration
v0 by Vercel
Great for quickly visualizing component ideas before building
What I've Learned
Prompting is a Design Skill
The quality of AI output directly reflects the clarity of input.
"Make me a login page" "Create a login page for elderly users with high contrast, 18px min font, single-column layout, and clear error states" Speed Changes What's Possible
When building takes hours instead of weeks, you can chase ideas that seemed too risky before.
AI Amplifies, Not Replaces
Every project still required design decisions: What problem matters? What flow feels right? AI handled implementation; I owned the intent.
Currently Exploring
Work in progress. Ideas taking shape.
AI Agents for Design
Building autonomous assistants for design system maintenance
Design-to-Code Workflows
Figma to production code with AI translation
AI for Accessibility
Automated accessibility checking and fix suggestions
Have a project idea? Let's chat.