# Notes on my work (draft)

Niyora, CPR Finder and Life, one topic at a time: what I did, and what I learned along the way. This is a working draft, and I’m still filling it in.

## Niyora: How a calm button turned into an AI companion for PMS

**Role:** Founder · Designer · Builder · Growth · **Timeline:** April 2026 to now · **Platforms:** Mac, Windows, iPhone, web · **Built with:** Claude Code, Tauri, Expo, React Native, Gemini

Niyora helps women with PMS and PMDD calm down, see the moment a different way, and figure out what to say next. I started it in April 2026 and I build it, from research to release. Here’s what I learned, one topic at a time.

- App Store: https://apps.apple.com/no/app/niyora-pms-mood-relief/id6764561688?l=nb

- niyora.com: https://niyora.com

### From a calm button to PMS first

Knowing when to stop is hard, and hardest during PMS. So in April 2026, the first thing I built was a little Mac app that tells you to stop and breathe.

#### Feelings that get bigger than the moment

During PMS, a feeling can get way bigger than the moment.

It comes out at the people you love, at work, or in a text you can’t take back. Right then, she needs to feel safe, feel in control, and respond instead of react.

Period apps asked me to log how I felt. I already knew how I felt. What I wanted was the next sentence, and nobody had one.

I built what I needed and couldn’t find.

#### A calm button in the menu bar

- In April I shipped a macOS menu-bar app that notices passive work signals and offers 60 seconds of breathing.
- In May I ported it to Windows, and dropped the fancy notification buttons so it could ship sooner.

![Niyora's Mac app open from the menu bar: a glowing pink orb above Box Breath, calms under pressure in 65 seconds, with a Begin button](https://neha-prasad.com/niyora/mac-app.jpg)
*April 2026: Niyora lived in the menu bar.*

#### I cut my favorite feature

If breathing helps, can I prove it? In May I built an Apple Watch companion to measure stress from your heart, and it was the most exciting thing I’d built.

- I built a Swift HealthKit companion app that paired with the Mac over the local network.
- I pushed for a full watchOS build to test it properly. One test on June 14 came back with 255 heart-rate samples and zero usable variability.
- The catch was simple. The watch only reads heart rate variability when you’re still. So I got clean numbers from the calm I was already in, and nothing from the fight.
- So I cut it. The watch couldn’t back up the promise, and a wellness app that guesses is worse than one that’s honest.

**Apple Watch test, June 14**
- Heart-rate samples: 255
- Usable variability readings: 0
The test that ended the feature.

#### Going PMS first

On June 20 the breathing app went live on the App Store. Two days earlier, I’d already decided what Niyora should really be: a companion for the bad moment.

- I pointed the whole product at PMS and PMDD, the moment when a feeling is bigger than the situation.
- In early July I shipped a PMS version: calm in a minute, cycle awareness, and still no sign-in.

![Niyora July version: a screen offering calm during PMS](https://neha-prasad.com/niyora/calm.jpg)
![Niyora July version: a screen to predict and prepare for the PMS days](https://neha-prasad.com/niyora/prepare.jpg)
![Niyora July version: a list of 30 ways to feel better](https://neha-prasad.com/niyora/feel-better.jpg)
*July 2026: the PMS-first version.*

#### What I learned
- Cut what you can’t prove, even if it’s your favorite.
- The two ideas I loved most, the watch and the on-phone model, both died. Letting them go was most of the work.

### 1,800 papers graded before I built more

I didn’t want activities that only sounded nice, so in July I built a research engine and read the evidence first.

- **1,800**: Papers collected and graded by study type, July
- **2,644**: Papers screened into the content bank
- **6,700+**: Plain-English paper summaries on niyora.com

#### A research engine and a reading cockpit

- In July I pulled about 1,800 papers from Europe PMC into a reading cockpit I built. It graded each one by study type, so the strongest studies showed up first.
- I set a rule: only findings with solid evidence show up in the app as fact.

[Video: The reading cockpit I built. About 1,800 papers, tagged and graded, one card at a time.](https://neha-prasad.com/work/niyora/research-cockpit-clean.mp4)

#### A content bank of eight stations

Then, with AI’s help, I screened 2,644 papers into a content bank: about 640 findings in eight stations, from understanding your body to relationships. Two stations came back thin, so I ran a second search just for them. Every fact in the app goes back to real research.

[Video: The content bank. Every finding sorted into the eight places it belongs.](https://neha-prasad.com/work/niyora/content-bank-clean.mp4)

![A Niyora quick fact: PMS comes from a brain sensitive to the hormone drop](https://neha-prasad.com/home/niyora-fact.jpg)
![PMS check result: your PMS reads as moderate on a scale from mild to PMDD, with movement, emotions, sleep and food ranked by how much they move your PMS, and a link to see the science](https://neha-prasad.com/niyora/pms-check.jpg)
*The research in the app: a quick fact, and the PMS check with a link to the science.*

By November, the research library on niyora.com held more than 6,700 papers in plain English, on the library engine Prasad Subrahmanya built.

#### What the research showed

The finding that changed how I saw my own month: women who feel awful before their period have the same hormone levels as women who feel fine. What’s different is how strongly the brain reacts to the drop.

One pattern stood out. The relief that lasts comes from learning to handle the emotion itself, so Niyora became a practice for that moment.

> See the hard days coming, catch the stress, help her feel safe again, check that it worked, and learn what works for her.
>
> The vision, one loop

#### What nobody has studied

In August I measured the whole field with 31 searches. There were 168 papers on the genetics of premenstrual symptoms, and zero trials. There was thirteen times more research on cycles in animals than on premenstrual self-harm. I published it as a gaps map on niyora.com.

#### What I learned
- When a field is this thin, confidence is the warning sign. The less evidence there was, the more careful the writing got.
- I didn’t need a fund or anyone’s permission to start. I could read the evidence myself.

### Designing Moon: I tested the AI first

Private by default meant trying a model that runs on the phone. So in late July I fine-tuned one and tested it before deciding.

#### I tried AI on the phone first

I fine-tuned a small open model on a few thousand practice examples and ran it on my own iPhone. It was good at reflecting feelings back, and weak at suggestions.

It finally ran on my iPhone on July 27, after a full day of debugging. The model itself was fine. Every fix was in how the app loaded it.

**Fine-tuned model on my iPhone, July 27**
- Good reflections: 77%
- Good suggestions: 26%
Why the on-device model didn't ship.

#### What the test shaped

That finding shaped the whole design. Moon listens and reflects first, and its suggestions stay grounded in the [research](https://neha-prasad.com/draft#niyora-research).

A reflection she can’t act on is just a nice sentence. Suggestions are the half that has to work.

- For version one I moved Moon to Gemini through Firebase. Moon’s home screen shipped on July 31.
- Small model, funny surprise: the word “moon” in a prompt made it write poetry. Now a test fails if that ever sneaks back in.

Moon never pretends to be human. It never calls itself a person, and there’s no endless chat to keep her typing.

[Video: Designing in Figma: the emotion work taking shape before it became Moon.](https://neha-prasad.com/work/niyora/app-design-clean.mp4)

#### Reflect, Regulate, Respond

Moon went live on August 20 in three steps: think about what happened, calm down, then choose what to say.

![Reflect screen asking what this feeling might be pointing to](https://neha-prasad.com/niyora/v3-reflect.jpg)
![Regulate screen suggesting activities to try before you respond](https://neha-prasad.com/niyora/v3-regulate.jpg)
![Respond screen with an editable draft message and options to make it shorter, softer or more direct](https://neha-prasad.com/niyora/v3-respond.jpg)
*Moon in August: think about what happened, calm down, then choose what to say.*

#### What I learned
- Test the model before you design around it. Its weak spot told me what Moon should be.

### Swipe: keep what fits

Moon offers other ways to see a hard moment. Testers showed me that she has to be the one who decides which ones fit.

#### Swipe almost didn’t happen

In June I prototyped a calm swipe deck in plain HTML and built it into the PMS version. On my phone, swiping felt like too much work in a bad moment, so a simple grid of activities won.

#### One in five was enough to lose trust

Testers taught me the sharpest lesson. If Moon shows five ways to see a moment and one feels a little off, people don’t shrug and read the other four. They stop trusting all of it, and they stop opening the app.

That middle ground was the hardest thing to find in the whole build. Moon had to say something worth reading without reaching too far. What finally fixed it was handing her the options and letting her pick.

So now she chooses: keep what fits, swipe away the rest, and change the feeling Moon picked.

![A Moon card over a night sea: another way to see it, from psychology, with the feeling dismissed and a link to change it, plus keep and discard buttons](https://neha-prasad.com/niyora/moon-card.jpg)
*Another way to see it. Keep it, or swipe it away.*

> Love the design, the moon is beautiful.
>
> An early user, September

That same early user thought the moon was a “pretty white circle,” and a scale looked like a slider she had to drag. I fixed both that evening. The moon got a caption, and the scale got split into three parts.

#### How the deck changed

In October Swipe came back, this time for perspectives. One big card at a time. A heart keeps it, a cross lets it go.

There’s no progress bar and no finish line. She stops whenever she likes and picks up where she left off. Swipe is there to help her feel better. Keeping her swiping was never the goal.

#### What I learned
- When in doubt, let users decide. One wrong suggestion in five is enough to lose trust, and choice wins it back.
- Letting her pick fixed trust faster than any prompt I wrote.

### Privacy as the foundation

Moon talks to women on their hardest days, so privacy shaped the architecture before it shaped a single screen.

- **0**: Accounts or analytics in the app
- **2,954**: Automated tests in the iPhone app, October

#### Zero tracking from day one

From the first Mac app in April, I chose zero tracking, local-only storage and no accounts, and accepted that I’d be blind on usage. Trust is the product.

I’d rather talk to women, again and again. So I invite women to lunch, we talk about PMS and their day, and we try Niyora together.

I build it myself with Claude Code and Expo. By October the iPhone app had 2,954 automated tests.

#### Privacy in the architecture

When I built Moon in August, privacy came first.

Privacy and running on AI turned out to be one decision. With everything encrypted on her phone, nobody could catch a mistake by hand. So the AI had to be good enough to run alone, and that made the product better.

- I said no to a backend and no to accounts. History is encrypted on the phone, and there’s no analytics SDK in the app.
- Names, emails and phone numbers get taken out of what she types before it reaches the AI, and crisis moments go to a helpline.
- Notifications never name the other person, because he might see her lock screen.
- I caught a bug by feel: the answers didn’t feel like AI. A setting had never made it into the release builds, so now every build gets checked before it ships.

> The ‘you are safe’ is a nice touch, as I was thinking that straight away.
>
> An early user, September

![Your patterns in Niyora: feelings and topics in one place](https://neha-prasad.com/home/niyora-patterns.jpg)
*Your patterns, kept on your phone.*

#### Consent she can take back

- I removed the last bit of tracking in October. I never wanted her to feel questioned about her safety.
- When Apple asked for clearer AI consent, I designed a screen that names Google and lets her take it back at any time.

#### Safety before Premium

A mental health app should finish its safety work before it takes anyone’s money. That’s the order I picked.

Moon went live on August 20. Premium waited until September, after the safety work.

Later I made Home and the downloadable report free for everyone, so Premium is just Moon.

#### What I learned
- Ship the safety before the price.

### Built for search

The research bank was too good to live only inside the app, so starting in August I made it the way women find Niyora.

- **8,000**: Pages on niyora.com, from one research bank
- **3**: App Store pages, one per reason women search

#### One research bank, published three ways

- In August I built the SEO layer: research pages and a gaps map on niyora.com, plus a Markdown copy of every page and llms.txt for AI search.
- In October Prasad Subrahmanya joined and built the page engine that grew the site to about 8,000 pages.

1. **One research bank**: Graded papers, symptoms, what helps and cycle science, all in one place.
2. **Published three ways**: Research pages and a gaps map on niyora.com. A Markdown copy of every page plus llms.txt, so AI assistants can read and cite it. App Store pages for each lens: PMS, ADHD and perimenopause.
3. **Search brings her to Moon**: On Google, in AI answers and in the App Store.

#### One App Store page per reason women search

I rewrote the App Store page around the question women actually ask: “Is it me, or is it PMS?”

On October 10 I renamed the app “Niyora: PMS & PMDD Moods,” with that question as the subtitle. Then I built pages for the ADHD and perimenopause lenses that open straight into them.

![App Store page for PMS: why am I so angry? Say it out loud, Moon listens.](https://neha-prasad.com/home/niyora-lens-pms.jpg)
![App Store page for ADHD: can't stop replaying it? Say it out loud, Moon listens.](https://neha-prasad.com/home/niyora-lens-adhd.jpg)
![App Store page for perimenopause: don't feel like yourself? Say it out loud, Moon listens.](https://neha-prasad.com/home/niyora-lens-peri.jpg)
*One app, a way in for each lens.*

#### What the numbers say so far

It’s early, and the numbers are honest: search works, and social hasn’t yet.

- Every download so far is organic. I haven’t spent anything on ads.
- App Store search brought in more than half of all downloads. I tried Reddit, LinkedIn and Instagram on the side, and they brought a handful of visits each.
- I’m still waiting on the first real review. That’s the next problem to solve, and the next quarter goes into search and into talking to the women using it.

#### What I learned
- For a private topic, search beats social. People look for answers alone, late at night.
- Explaining PMS honestly is the product, and it’s also how women find it.

### Moon got a voice

In October, a friend told me she’d rather talk than swipe. That became Moon’s voice mode.

#### From swiping to talking

- I started building it the same day. The first build connected but couldn’t hear me. The next one, on October 14, worked end to end, and I had my first real talk with Moon.
- Moon’s voice mode runs on Gemini Live, with Prasad Subrahmanya building the voice engine.
- Then I tested it on my own bad days. Moon asked too much. In one talk it asked if anything else was on my mind four more times after I’d said no. In another, it barely let me finish a sentence.
- I went back to how counselors are trained to listen, and redesigned Moon’s voice to reflect more and ask less.
- Now Moon waits until she’s done, asks one question at a time, and she decides when the talk ends.
- Every talk ends with the same reminder: Moon is an AI, and it’s there to help her with the people in her life.

![Moon listening in voice mode: tell me what happened, and how you're feeling](https://neha-prasad.com/niyora/moon-voice.jpg)
*Tell Moon what happened, out loud.*

#### Where Niyora is now

From a breathing app in June to version 3.4 in November, I shipped in small steps, each one shaped by what the last one taught me.

1. **1.0 · June 20**: The breathing app, live on the App Store.
2. **2.0 · Early July**: PMS first, with 30 ways to feel better.
3. **3.0 · August 20**: Moon: reflect, regulate, respond.
4. **3.2 · September**: Premium, after the safety work.
5. **3.3 · October**: Period days from Apple Health, and photos as context.
6. **3.4 · November 1**: Moon's voice mode, and lenses for ADHD and perimenopause.
7. **Next · Android**: In progress, from the same code as the iPhone app.

- Version 3.4 shipped on November 1, with Moon’s voice mode switched on next to Swipe, and lenses for ADHD and perimenopause alongside PMS.
- Talking is free for everyone, 15 minutes a month. Her voice goes to the AI as she speaks, and Niyora doesn’t record it.
- Since October, Niyora reads period days from Apple Health instead of asking, and she can add a photo or screenshot to show Moon what happened.
- An Android version is in progress, built from the same code as the iPhone app.

#### Watch a design session

[Video: A full design session, unedited. Ten minutes of the thing being built.](https://neha-prasad.com/work/niyora/building-games.mp4)

#### What I learned
- Rules are tools. When Moon followed the flow too closely, it forgot what she was actually saying.
- Listen more than you ask. Moon got better when it reflected more and asked less.

## CPR Finder: Getting 2.9M health workers into the right CPR class

**Role:** Sole designer · Builder · Strategy · Data · **Company:** Laerdal Medical × American Heart Association · **Years:** 2024 to now · **Tools:** Figma, Figma AI, Claude Code, PostHog, Playwright

I’m the only designer on CPR Finder, for Laerdal Medical and the American Heart Association. I research it, design it, write the code and measure how it does live. A lot happens on a product like this, so here it is one topic at a time.

### Getting a CPR card was a maze, so I took the course myself

Every health worker in the US needs a CPR card to work. When I went through it myself, almost every step was confusing.

- **52.4K**: Survey answers from learners, across 12 organizations
- **5.7/10**: Self-service score across three learner journeys
- **2**: US trips: the full course in California, then the skills check in New York

#### What I ran into, start to finish

- Course names and prices were confusing, so it was hard to know what to buy.
- I got three emails, and none of them linked to the course.
- There was a two hour wait before I could start the course, and nobody told me.
- The online course took about two hours, which I didn’t expect, and it had so many buttons I wasn’t sure what to click.
- At the end, the certificate was hard to find, and the next steps weren’t clear.

On the way to booking, a learner meets 26 of our terms: HeartCode, Part 1, skills check, CPR Verification Station and more. Most of them aren’t words learners use.

On top of that, the market is full of confusing and fake courses. Employers had a hard time finding real providers, and learners lost money on fake certificates.

#### I became a learner first

Before designing anything, I took the AHA course start to finish.

- In February, three of us went to California and visited three training sites, in Los Angeles, San Diego and San Francisco. At each one I booked, paid, did the online course and then the skills check, just like a real learner.
- In September I went back, this time to New York, and did only the skills part, using the online certificate I already had.
- I filmed the whole thing so the team could feel it too, and designed CPR Finder around how it really works.

[Video: My own learner journey, from online course to skills check, filmed for the team.](https://neha-prasad.com/work/laerdal/BLS_Neha.mp4)

> I failed once, and I was already struggling. Yet we only call a learner ‘struggling’ after 4 failures.
>
> Me, in the film

The second time felt different. The site checked my online certificate right away. My phone told me where to go and what to do next. I failed the switch from the adult to the infant manikin, and I was tired. A coach came over, muted the voice prompts, and coached me from my own data instead of my posture. Then confetti, and my eCard by email. The one snag: claiming it asked for my details again.

#### 52.4K answers in learners’ own words

I read 52.4K survey answers from 12 organizations, plus Reddit and nursing forums, to hear the frustrations in learners’ own words.

**Who the learners are, from 52.4K survey answers**
- Hospital or clinic provider: 64%
- Student in training: 20%
- EMS or prehospital: 7%
- Other: 9%

#### Where we started

Then I scored 9 steps of the self-service journey, from finding a provider I could trust to getting the certificate. I rated my own experience at each step, then rated other people’s experiences too. It came out to **5.7 out of 10**.

The score showed where to start: finding and booking a course. So we started with the [AI course finder](https://neha-prasad.com/draft#cpr-finder-course-finder).

#### What I learned
- Take the course yourself first. Find the gaps, rate them, explain them, then research who needs which course.
- Filming my own course let the whole team feel what learners feel.

### An AI course finder with guardrails

Learners had to guess between BLS, ACLS, PALS and Heartsaver. I built an AI tool that asks a few questions and recommends the right one.

#### The biggest pain: picking the wrong course

Employers say “get CPR certified.” Then learners face BLS, ACLS, PALS and Heartsaver, first time or renewal, online or in person. So they guess.

A wrong guess means a refund, a rebooking, a support ticket, or a learner who just gives up.

Career switchers often pick the course with the longest name, because it sounds safest. That’s usually PALS, the hardest one to pass. One learner spent their last $100 on a course, and then their employer didn’t accept that format. The training site had to refund them.

I brought the problem to the team with the evidence and got everyone on board to fix it. We set one goal: every learner books the right course the first time. Right course, more finished bookings.

#### Which job needs which course

Nurses usually know they need BLS. The people who guess are everyone else. Electricians and park staff, for example, often need BLS plus first aid, because their job can mean using a bag-valve mask.

I mapped US jobs by sector to the CPR course each one needs. That map feeds the course finder.

![My research board: US jobs grouped by sector, each colored by the CPR course it needs, from BLS and Heartsaver to ACLS and PALS](https://neha-prasad.com/home/cpr-jobs.jpg)
*Which sector, which course: my research board.*

#### How the course finder works

CPR Course Finder is RQI’s first AI tool for learners. It asks about your role and your job, then recommends the right AHA course.

My first version asked a lot: your role, what your job needs, whether you’d rather practice in a group or alone, your budget and your time. I wanted it to feel like ordering coffee your way. Then a product lead at RQI Partners told me the 3 questions she asks every learner on the phone. I rebuilt the finder around those.

1. **The learner answers a few simple questions**: What is your role? Do you need the card for work? Do you respond to emergencies at work? Do you need first aid too?
2. **AI reads the answers, inside guardrails**: Guardrails keep every answer to a real AHA course, in the right format.
3. **One clear answer**: For example: BLS plus first aid, self-guided at a CPR Verification Station, with about 45 to 60 minutes online.

#### Getting it ready to launch

- I took it from problem to working tool myself: I researched it, designed it and built it, then had the developers test it.
- I’m working with Laerdal’s AI and Data team on guardrails, testing and data masking, so it meets our privacy rules before it goes live.
- It’s built to sit where learners start. The AHA team is interested in adding it to Atlas, the course booking flow on the AHA’s own website.
- My pitch: if the AHA embeds it, we embed it on CPR Finder, and distributors can embed it too. Then we all speak the same language, and learners can move between the AHA’s site and ours.

The course finder is getting ready to launch, and I’m leading it.

#### What I learned
- Bring the evidence. Showing the team the pain in learners’ own words got everyone on board to fix it.
- The best questions come from the people on the phone with learners every day. Theirs beat my long list.

### Live in 9 days, then Book & Pay

First the new site, live 9 days after my first session with Branding. Then Book & Pay, so learners can book and pay online without a phone call.

- **9 days**: From rebrand to live site, with 4 days of user testing
- **+3%**: Lift in conversion from the new landing page to the Map
- **85%**: Faster component library with Figma AI

#### The new site, live in 9 days

The rebranded CPR Finder went live 9 days after my first working session with Branding, July 20 to 29, with 4 days of user testing in the middle.

It started with two 3-hour working sessions with Branding, on July 20 and 23. Two days later the new look was live as an experiment. After 4 days of user testing, the site launched on July 29.

- I designed and built the landing page, About, Courses, FAQ and the path to the Map myself, and took them all the way to the A/B test. That let the developers stay focused on Book & Pay.
- I added the motion myself. I came up with the animations, got them approved and built them, so changes on screen read clearly.
- The new path from the landing page to the Map lifted conversion by 3%. Learners who know what they want book more often.

![CPR Finder's Find a CPR Verification Station page: a search field and a list of nearby training sites with distances on the left, and a map with location pins on the right.](https://neha-prasad.com/work/laerdal/cpr-finder-find.png)
*Find a station: nearby sites in a list, next to the map.*

#### Book & Pay: booking without a phone call

I designed Book & Pay and its component library, then built it with the developers, so learners can book and pay online.

Before, learners found a site on CPR Finder and then left to book somewhere else. Lots of US learners still book by phone, and many sites don’t show prices, so people call around to compare. With Book & Pay they book and pay on CPR Finder, then go straight into the online course.

Learners pick a course, pick a date and time, pay, and get clear next steps: the online course, the skills session and the e-card.

![A training site's booking page with photos, a How it works section and a Reserve your spot panel listing BLS, ACLS, PALS and Heartsaver courses with durations and price.](https://neha-prasad.com/work/laerdal/cpr-finder-reserve.png)
![Booking confirmed page with an order summary and three next steps: the online course, the hands-on skills session and the e-card.](https://neha-prasad.com/work/laerdal/cpr-finder-confirm.png)
*Booking a course, then the confirmation with next steps.*

#### A component library in 6 hours

I built the booking flow’s component library with Figma AI in about 6 hours. By hand, that takes me 35 to 45. Now designers and developers build from the same components and tokens.

**Hours to build the booking flow's component library**
- By hand: 35 to 45 hours
- With Figma AI: About 6 hours
By hand is how long the same components take me in Figma. I built them both ways and compared. By hand is drawn at the middle, 40 hours. About 85% faster.

I demoed the new design to a training provider, and this is what they told me.

> This design is 10x better than our current tool. With a few additional features we’ll move to this product.
>
> Distributor, USA

Book & Pay launches in November 2026.

#### What I learned
- Speed and quality can go together. Two short workshops, and the new brand was live in days, bringing in more trust and more bookings.
- Split the work so everyone moves. I built the site pages myself so the developers could stay focused on Book & Pay.

### One AI workflow, from research to pull request

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.

### SEO and AI search, set up from scratch

A marketplace lives on traffic, and nobody on the team had done SEO or AI search hands-on. So I set up both myself.

- **200**: Booking pages across 13 distributors
- **2 min**: For a distributor to approve or edit a draft

#### Setting up SEO and AI search

- I set up SEO and AI search on the site, and I keep coaching the team on both.
- I built an AI tool and workflow for RQI Partners’ Marketing team that auto-fills keywords for the location pages.
- I wrote the writing rules for distributors’ course pages, plus an AI check that flags anything off before a page goes live.

#### How a station page gets written

Every station needed its own page, written by the distributor. Writing hundreds of unique pages wasn’t realistic for them, and checking every page for rule breaks was too much for our team. So I had an idea and built it. Distributors save time, our SEO gets better, and learners get honest pages.

1. **200 booking pages across 13 distributors**: Every page needs its own content, because copied text doesn't rank.
2. **Crawl, pull APIs, find keywords**: From the distributor's site, reviews, Yelp, Google Places and Search Console.
3. **AI drafts the page**: A unique description and FAQs, in Marketing's voice.
4. **The distributor approves**: Approving or editing a draft takes about 2 minutes.
5. **AI rule check**: Flags anything that breaks the writing rules.
6. **Publish**: Each page goes live with its own content, ready to rank and bring in bookings.

![A training site's booking page with photos, a description and a Reserve your spot panel](https://neha-prasad.com/home/cpr-station.jpg)
*A booking page with its own description.*

#### What I learned
- Copied text doesn’t rank. Every page needs its own words, and AI can write them as long as people approve them.

### List view first: +7% mobile conversion

I own CPR Finder’s product analytics in PostHog. It’s the team’s feedback loop: track, find, test and measure, all live.

- **+7%**: Mobile conversion after making the list view the default
- **50%**: Of visitors click through to a distributor to book

#### One loop, start to finish

1. **Track**: How mobile learners get from search to booking.
2. **Find**: Mobile learners converted better from the list view than from the map.
3. **Test**: We made the list the default on mobile.
4. **Measure**: Mobile conversion went up 7%.

With the list first and the map second, mobile converted about as well as desktop.

I also A/B test the landing page itself, like which kind of photo gets learners to search for a station.

![CPR Finder on a phone: nearby training sites in a list, with a Book now button](https://neha-prasad.com/home/cpr-list.jpg)
*The list view, now the default on mobile.*

#### Measuring what learners actually do

I pushed back on learner targets that didn’t match how people actually behave. Now I track the share of visitors who click through to a distributor to book: about 50%.

#### Data for every team

- I build dashboards for everyone from Marketing to impact managers, each showing the numbers behind that team’s decisions.
- I get data on learners who struggle from our data team and go through it with RQI Partners’ impact managers, so distributors can see where learners get stuck.
- I ran a [data quiz](https://neha-prasad.com/passion/data-quiz) on learner behavior, personas and the US market, so the whole team decides from the same picture.

#### What I learned
- Taste vs test. An A/B test gets us to a decision faster than going back and forth on opinions.
- I judge my work by how it does live, and I let the data change my decisions.

### Partners and team

CPR Finder sits between Laerdal, the AHA, RQI Partners and dozens of distributors, so I put real time into the relationships and the team behind it.

#### Who I work with

I work with a product lead, developers, Marketing and Laerdal’s AI and Data team, plus partners at the AHA, RQI Partners and US training distributors. A change for one moves bookings and revenue for the others.

#### Partners

- In September I spent two full days on site with two distributors, sitting with them for hours to see how learners call, book and show up. I also spent five hours with the AHA’s product team.
- I coach RQI Partners’ Marketing team on positioning, SEO and AI search, based on learner data.
- I bring feedback from distributors and impact managers into the product, as research.
- I join the AHA’s Atlas design conversations with our learner data and what distributors tell us, to make things better for the learner.

#### A safe place to grow

Trust is what lets a team try new things, and I put a lot of time into building it.

- I ran the workshops where the team got aligned on the learner journey, roadmap and vision. In one of them we wrote our vision: to empower lifesavers to seamlessly find, book and complete self-guided CPR training, anywhere, anytime.
- I saw people didn’t share the same picture of what CPR Finder was. So, on my own, I made short videos to explain it.
- In India, I ran a workshop on how Indian, Norwegian and American coworkers communicate differently, and how each can play to their strengths.
- I wrote the team’s onboarding docs on my own, and I do 1:1 onboarding for new people, developers included. Nobody asked me to. New teammates, developers too, told me it helped.
- As UX Hive lead and Laerdal’s AI champion, I run workshops on AI, [motion design](https://neha-prasad.com/passion/motion-design) and [UX writing](https://neha-prasad.com/passion/content-writing) for designers and product teams.

> Neha doesn’t just talk about ideas, she makes things happen. She is a true go-getter who is consistently focused on creating real impact.
>
> Ola Helland, Design Director, Data & AI, Laerdal Medical

#### What’s next

Next, CPR Finder goes from helping learners find a course to helping them book it, with every step measured.

- Book & Pay launches in November 2026, and the course finder is getting ready to launch. Then I measure the whole journey, from search to certificate.
- A UI designer is joining, so I’m moving my focus to UX, user research, strategy and AI product work.

[Video: The vision film I made to show where CPR Finder is headed. Sam, the learner in it, is an AI-generated persona.](https://neha-prasad.com/work/laerdal/Sam_the_learner.mp4)

#### What I learned
- Relationships build trust, and trust makes a safe place to grow.

## Life: Growing Life from 100K to 1M women

**Role:** Founding Product Designer · **Company:** Infinity Games · **Years:** 2022 to 2024 · **Platforms:** iPhone and Android

I was the founding product designer on Life, a women’s health app for iPhone and Android, from 2022 to 2024. It grew from 100K to 1M women, with 120K using it every month across the US, Europe, South America and Asia. A lot happened in two years, so here it is one topic at a time.

### A symptom quiz built from 200+ interviews

Every feature on Life started with something we kept hearing in interviews. The symptom quiz and the AI assistant came straight from them.

#### What women kept telling us

Most women don’t know how much of what they feel is normal, or worth doing something about.

I ran 200+ interviews and read threads about stress, mood and cycles. They showed us where our words didn’t land, and what women actually wanted to know.

Before Life, I worked in women’s wards at a hospital. I taught women how their bodies work, and I interviewed them and wrote down the changes they went through in menopause. With that background, I also wrote the logic behind the app’s health tips.

I turned what I heard into decisions about what to track, how to personalize it, and how to build a daily habit.

#### A quiz that never sounds clinical

I built a quiz that asks about your symptoms in plain words and turns your answers into a warm personality result.

It opens with one plain question: which changes do you notice before your period? Mood, body, energy, or none. Pick as many as you like.

![Life symptom assessment ending in a personality result](https://neha-prasad.com/photos/life-assessment.png)
*From symptom check to personality result.*

At the end you get a character. One result says “You are the body listener,” with scores for energy, mood and stress recovery, and a meter that shows how much your period affects your days.

#### An AI assistant that turns knowing into doing

An AI feature reads where you are in your cycle and shows a daily tip.

In chat, it suggests four things to try for that phase, across food, mind, movement and relationships, so knowing turns into doing.

The assistant is called Livia. She says it up front: she’s not a doctor, but she knows your cycle logs and can help.

In the luteal phase, Livia might suggest sweet potato and spinach for magnesium, a few minutes of gratitude journaling, 20 minutes of gentle yoga, and pausing before you react.

![Life AI chat suggesting actions for the current cycle phase](https://neha-prasad.com/photos/life-ai-chat.png)
*A daily tip, then four ideas per phase in chat.*

#### What I learned
- Every answer a woman gives should give her something back. Period apps that only collect data wear people out.

### Onboarding: trust first, data later

Life asked for really personal info before people trusted it. I rebuilt onboarding so every question earns its answer.

#### Where Life was when I joined

Life had 100K women using it, and revenue kept dropping month after month. A lot of women stopped showing up after a week.

From the outside the app looked fine, but the basics were broken. Four things stood out, and no new feature was going to fix them:

1. It asked for really personal info before people trusted it.
2. It didn’t show what it was good for in the first few seconds.
3. The paywall showed up before the value did.
4. New features had stopped moving any number that mattered.

I was the only designer on a lean team, so I fixed the basics in the order a woman runs into them, starting with her first minute.

#### A new order that earns every answer

The old flow asked for your cycle length, birth control and mood in the first 60 seconds. It felt like filling out a form at a clinic.

The new order:

1. Show what Life is for.
2. Ask only for the one thing we need right now.
3. Give the first reading, then ask for more.

Even the questions got softer. When we ask about cycle length, the screen says most cycles last 26 to 35 days, and it’s fine if yours is different.

[Video: The new onboarding, screen by screen.](https://neha-prasad.com/videos/onboarding.mp4)

#### Tested one step at a time

Mixpanel funnels showed where women dropped off, flow by flow. Every drop-off became one focused A/B test. I made small changes I could undo, one at a time.

Together with the pricing and [paywall tests](https://neha-prasad.com/draft#life-paywalls), this work brought **+40% paid growth on iOS**.

> Neha played a crucial role in the complete restructuring of the app and demonstrated a deep understanding of users’ needs. Her passion for addressing women’s health issues stood out.
>
> Filipe Cruz, Full-stack Developer, Life

#### What I learned
- Earn trust before you ask for data, especially personal data.

### A home screen that works like a daily forecast

Women open a period app with one question: what’s going on with me today? I turned the home screen into the answer.

#### A forecast for your day

I turned the home screen into a forecast: your emotional weather, a stress score and a countdown to your next cycle.

Every card answers the question a woman opens the app with: what’s going on with me today?

The cards talk like a friend. One said: “Still the will to overdo it. You’re pushing hard despite low energy. Try to recharge today.”

![Life daily feed showing today's emotional weather](https://neha-prasad.com/photos/Feed1.png)
![Life daily feed showing a stress score](https://neha-prasad.com/photos/Feed2.png)
![Life daily feed showing the cycle countdown](https://neha-prasad.com/photos/Feed3.png)
*The feed over three days: forecast, stress and cycle.*

It’s one of the 120+ screens I designed as the only designer on Life.

#### Seeing the hard days coming

Three out of four women go through emotional changes with PMS.

The emotional forecast lets them see it coming, phase by phase, so they feel ready.

Every phase gets its own weather. Late luteal is a grey, cloudy sky that says “High chance of PMS, take care.” Mid luteal is a misty forest: “Be gentle with yourself.” Ovulation is a sunset on a lake: “Go out and socialise.”

Each of the six phases explains what your hormones are doing in one line, shows a hormone chart, and gives four tips. In the late luteal phase, one of them is: let them know it’s PMS time.

![Emotional forecast showing expected moods for each cycle phase](https://neha-prasad.com/photos/life-emotional-forecast.png)
*The emotional forecast, phase by phase.*

#### What women said

I read every review. The love and the survey answers both shaped what we built next.

We asked women which part of the feed helped most. The period history graph came first. Then the emotional forecast, then the stress, energy and health scores, then the daily streak.

> It tracks my stuff accurately and gives wonderful insight.
>
> App Store review

> I think what you have right now is amazing. Just my opinion, keep it up.
>
> App Store review

> You are super. I have been with you for 5+ years.
>
> App Store review

#### What I learned
- People come back for a reason. A forecast for today is a good one.
- People come back to see their own story. In our survey, the period history graph beat everything else on the feed.

### 200+ paywall tests, and the one that won

The old paywall showed up before the value did. I treated every paywall like an experiment, and ran over 200 of them.

- **+23%**: Revenue from one paywall
- **+40%**: Paid growth on iOS
- **20×**: In-app purchases
- **37%**: Of monthly users on paid plans

#### How I tested: watch, test, tweak

I treated every important flow like something I could test, watch and tweak, one experiment at a time.

Every flow ran on the same loop: guess, test, learn, change. We ran it on acquisition, onboarding, engagement and the paywall.

1. **Watch what people do**: Mixpanel funnels showed where women dropped off, flow by flow.
2. **Test one change**: Every drop-off became one focused A/B test, from onboarding steps to paywall copy.
3. **Nail the first impression**: App Store tests tried out screenshots and messages before anyone even installed.
4. **Listen**: 200+ interviews and threads about stress, mood and cycles showed us where our words didn't land.

This loop kept us honest. A change only shipped if it moved a number, and the interviews told me why it moved. I built the winning ideas into patterns we could reuse and A/B test again.

#### 200+ versions, each one a small bet

The paywalls that won were rarely the prettiest. They said the value out loud, right when a woman was already feeling it.

I tested headlines, prices, plan options, buttons, illustrations, even punctuation. Each version was a small bet I could measure.

I planned a paywall for each step of the journey and priced each one to the moment. Onboarding offered a free trial. If she closed it, she saw a lower price. The home screen ran a banner, generic or seasonal. Locked premium features had their own paywall. When she came back to the app, the price matched that visit.

![Life onboarding paywall offering a personalized plan](https://neha-prasad.com/photos/paywall-onboarding.png)
![Life in-app upgrade screen listing premium benefits](https://neha-prasad.com/photos/paywall-upgrade.png)
*Two of the versions: the onboarding paywall, which converted the most, and the in-app upgrade.*

#### The winner: Easter Surprise

One paywall beat every other version on revenue: Easter Surprise bumped it 23%.

Four things made it work:

- A simple, direct message.
- It went live over Easter weekend.
- An image that felt real.
- A short offer with a deadline.

It offered half price for good, but only for a few days. If you tried to close it, a sad bunny asked: “Are you sure you don’t want a discount?”

![Easter Surprise paywall with a time-limited discount on Life Premium](https://neha-prasad.com/photos/life-paywall-easter-v2.png)
*Easter Surprise: +23% revenue.*

Together with the [onboarding](https://neha-prasad.com/draft#life-onboarding) and pricing tests, the paywall work brought **+40% paid growth on iOS** and **20× in-app purchases**. **37%** of monthly users were on paid plans.

#### What I learned
- Better-looking designs didn’t always win. In several tests, the less polished one beat the clean one. Paywalls win on clarity and timing.
- Small bets you can measure add up. 200 tests taught me more than any big redesign.

### A community women come back to

People need a reason to come back. I designed a safe space for the honest conversations women rarely get to have.

- **+43%**: Engagement
- **+18%**: Retention
- **+12%**: Revenue

#### What I built

A lot of women stopped using Life after a week. I built the community from scratch to give them a reason to come back.

In the app it’s called Safe space. It has two tabs: HerStory for talking, and Resources for learning.

- Discussions, polls and stories, moderated with AI to keep it safe.
- I grew it with content and wrote the articles myself.

![Life community home with discussions and polls](https://neha-prasad.com/photos/life-community-1.png)
![HerStory, where women share their experiences](https://neha-prasad.com/photos/community-herstory.png)
![Resources section with health articles](https://neha-prasad.com/photos/community-resources.png)
*Community home, HerStory and Resources.*

One discussion asked: “Anyone else feel marriage changed your confidence at work?” A poll asked which period product they’d recommend.

Resources started with quick questions like “Feeling hot even when it’s cold?”, plus real stories from women, including stories about menopause.

#### Staying close to the women using it

With my background in medicine, I did more than screens. I wrote the in-app articles, and I answered support tickets to stay close to real users.

#### What the community changed

**After the community launched**
- Engagement: +43%
- Retention: +18%
- Revenue: +12%

#### What I learned
- People come back for people who get it.

Live: https://neha-prasad.com/draft
