Inspiration

Nobody gets promoted for the idea they couldn't explain.

I've spent years as a mediator, Agile coach, and release train engineer on defense and space programs. I hold a law degree. None of that saved me in the moments where I knew the answer and struggle to deliver it under pressure. Credentials don't close that gap. It's a separate skill, and almost nobody practices it until the moment it counts.

So I built the thing that closes mine.

What it does

Sipra – Talk Smart is a 7–9 minute daily workout for the moments that decide things: the raise you ask for, the room you have to hold, the interview you get one shot at.

Each session follows the same arc:

  1. Choose a plan. Pick one of four goals, your time, and your level.
  2. Reset your state. Box breathing before you speak. A steady body makes a steady voice.
  3. Learn, then practice. Each drill teaches the technique first, then asks you to use it out loud. Once you know the technique, you can hide the teaching step.
  4. Record and play back. Hear yourself say it, and hear yourself improve over time.

Every recording stays in a private vault on your device. You can delete it at any time.

How we built it

  • App: Flutter/Dart, one codebase shipping to iPhone and iPad.
  • Auth: Firebase Authentication with Apple, Google, and email sign-in.
  • AI drills: Three of the five daily drills are generated fresh by Claude through a Firebase Cloud Function. They're prefetched before the session starts, so there's no loading pause between drills.
  • Monetization: RevenueCat manages a single Sipra Pro entitlement with two offerings:
    • Monthly: $6.99 with a 3-day free trial
    • Annual: $49.99

Against twelve months of the monthly plan, the annual plan saves

$$1 - \frac{49.99}{6.99 \times 12} \approx 40\%$$

The paywall appears at two natural moments: from the account screen, and right after a completed session, when the user has just felt the value.

  • Privacy by design: no ad SDK, no analytics, no third-party tracking. Recordings never leave the device.
  • Build partner: I'm a solo founder and wrote the app with Claude as my pair programmer.

Challenges we ran into

My first prototype wasn't showing real data. It showed Clarity: 62% and a four-day streak. It looked finished, but it measured nothing: the numbers were invented placeholders. Six weeks in, I tore them out and wrote the services that should have existed on day one. The focus bars now show progress for each of the areas covered after the drill.

The subscription identity bug. After a purchase, the app wasn't seeing the user's Pro entitlement / the RevenueCat customer didn't match the signed-in Firebase user. I worked it through with Rico, from RevenueCat's team and Claude until we found the cause in the app-side refresh and identity handling. One line on the fix, e.g. "We now log the Firebase UID into RevenueCat on sign-in and refresh customer info after purchase."

App Review. My first build was rejected under Guideline 2.1 because of a sign-in issue with the demo account. ****

Accomplishments that we're proud of

  • A real app, not a demo. Shipped to iPhone and iPad with live subscriptions.
  • Teach before you test. Every drill explains its technique before asking you to use it. You learn from it before you're asked to perform.
  • AI that can't break a session. If drill generation fails, a curated drill loads in its place. The user never sees an error, and the session always finishes.
  • Privacy as a feature. People will only practice their clumsiest first attempt if they know nobody else will hear it.

What we learned

  • AI needs a safety net. Generated content is a bonus, not a dependency. Fallbacks and prefetching made the AI layer invisible, which is how it should feel.
  • Fake metrics are debt. A placeholder number feels like progress and teaches you nothing. Measure something real, or show nothing.
  • Monetize at the moment of value. Users get through a free trial period before the paywall prompts. The paywall belongs right after a session, when the user has just felt the difference, not in front of the first drill.

What's next for Sipra – Talk Smart

  • Objective feedback signals. On-device speech analysis to count filler words and measure pause length. These will be one input scored against the framework, not the whole score.
  • Scenario packs. Drills for new managers (giving feedback, setting boundaries, saying no), job interviews, and salary negotiation.
  • More languages. Practice in the language you'll actually have the conversation in.
  • Research. A study with a neurodivergent-focused nonprofit to test whether structured, private rehearsal helps the people for whom spontaneous conversation is hardest.

Built With

Share this project:

Updates

Submission history