Speachr

Inspiration

I Have been a huge user of text to speech tools for a while, My parents specially my dad use to face this problem where he has to send a text or email to a Large Organization and I would have to correct it after he wrote the content in his own words. Now I Wisprflow is a app that solves this exact problem but It did not translate the text.

What it does

So to solve this specific problem I created Speachr. It turns raw messy speech into clear, structured and translated into desired language Text. It works with any app you just tap on the text box then tap the floating bubble and speak whatever you want our app will pick your language and transcribe , correct and format it fixing the grammar, punctuation and make it sound something you would actually send

How we built it

client - Kotlin + jetpack compose (Native android app only)
server - serverless on cloudflare workers JS
cloud - supabase for DB and auth, Cloudflare workers and workers AI
Open AI whisper - for speech to text
DeepSeek Ai - text enhancement

Challenges we ran into

I was a fullstack dev with years of experience with react and that is the reason why I started building the app with react native and realized that this needs way many android specific customizations.

Then I started learning kotlin and also found shipathon from their website.

Building a non generic UI fast with compose which I was using for the first time was a bit rough journey.
Translation that preserves intent, not just words. sometimes in a long text the ai looses the meaning of the context in text or translates it litterally.
Choosing models per step instead of one model for everything.

Accomplishments that we're proud of

Shipping a working end-to-end pipeline record → transcribe → correct → translate that feels like a single instant action to the user, not four separate steps.
Making translation a first class, core part of the product rather than an afterthought, solving a real gap I personally kept running into with existing tools.
Landing on a visual identity that feels distinct and polished rather than templated, on a hackathon timeline.

What we learned

Learned OOP, how android apps actually work and fell in love with kotlin along the way.
Practical tradeoffs between different LLM providers for different sub tasks, rather than defaulting to one model for the whole pipeline.