# One AI workflow, from research to pull request (draft)

CPR Finder · Laerdal Medical × AHA. Published 10 November 2026, updated 15 November 2026. Topics: design to code, AI, design engineering.

*This piece is a draft. I’m still filling it in.*

Speed matters on a new product, so I write the code myself, in the same codebase as the developers.

- **Top 4**: Code contributor, and the only designer on the list
- **6,370**: Lines of code, July to October

## How I got onto this team

I got onto this team by building something nobody asked for. I’d been making small AI tools for myself. Then I found a real problem at work: sales teams struggled to keep track of expensive demo units, and someone called around every week to find them. I built a tool and made a demo video. The video spread across the company, and when a team formed to build it, they picked me to lead the design. That path led me to CPR Finder.

## From research to pull request

Every step runs through one AI workflow, from the first brainstorm to the pull request.

1. **Brainstorm**: Research recordings and briefs.
2. **Workshop**: Miro.
3. **Design together**: Figma.
4. **Document**: Confluence.
5. **Plan tickets**: Jira.
6. **Test**: Playwright.
7. **Build and PR**: Review, then iterate.

## Real code, in the same repo

I design and build with Claude Code and Figma AI in the same codebase as the developers.

I don’t wait for someone to write a ticket. I find the gap, design the fix, build it, check what it changed, and keep going.

- I’m one of the top 4 code contributors to the CPR Finder repo, and the only designer on that list: 6,370 lines from July to October.
- I use the same pull request and code review process as everyone else. I open the PR, and a developer reviews it.
- I set the handoff standards: Confluence pages that cover UX, trade-offs and technical blockers. That made the path from design to production shorter.

Working this way, I can take a design to production in days. That’s how the new site went live in 9.

![VS Code open on the CPR Finder marketing site repository, with Claude Code in the side panel editing the CSS that aligns the landing page's audience card images.](https://neha-prasad.com/home/build.jpg)
*Claude Code in the CPR Finder repo, fixing image alignment on the landing page.*

## A QA system that tests like our learners

Small team, lots of flows, lots of screen sizes. So I gave our testing a memory and let it walk through the product as every kind of learner.

1. **A brain keeps the context**: Call recordings, docs and my notes after every meeting, all in one place.
2. **It reads Jira and GitHub**: So it knows what changed, what broke and what's still open.
3. **Playwright runs every flow as each type of learner**: On phone, tablet and desktop sizes.
4. **I design the fix in Figma**: Where a flow breaks, I design what it should do.
5. **Tickets with proof attached**: Each fix goes into Jira with what the test found.
6. **Work in parallel**: I take a few tickets, and the team works on the rest at the same time.

## What I learned

- I mean it when I own it. I find the gap and ship the fix.
- Work in the same repo as the developers. Same pull requests, same reviews, and a shorter path from design to production.

## Related

- [1,800 papers graded before I built more](https://neha-prasad.com/work/niyora/research.md)
- [Designing Moon: I tested the AI first](https://neha-prasad.com/work/niyora/designing-moon.md)
- [Swipe: keep what fits](https://neha-prasad.com/work/niyora/swipe.md)

Part of [CPR Finder](https://neha-prasad.com/work/cpr-finder.md) by Neha Prasad. Live: https://neha-prasad.com/work/cpr-finder/design-to-code
