About STEMBridge Speaks

I'm Aarish, a student based in Islamabad, Pakistan, and the founder of STEMBridge. STEMBridge started as a free STEM education initiative — the idea was simple: a lot of good learning material is locked behind subscriptions, and that shouldn't be true for students who genuinely want to learn.

STEMBridge Speaks came out of a personal frustration: I was learning German, and every app I tried either taught me grammar in isolation, or offered real conversation practice behind a $10-30/month paywall. I couldn't find a free tool that let me actually speak out loud, in a real situation, and get corrected the way a patient native speaker would correct a friend.

So I built one. It's still early — currently covering German, English, Spanish, Mandarin, Russian, and Hebrew, with more languages planned as the project grows. It's built and maintained by me, and every piece of feedback from real users has directly shaped what's been fixed and added.

How the correction actually works

When you speak, your audio gets transcribed, and the tutor reads it the way a person would — not against a rigid answer key. If what you said was understandable but a little off (wrong case ending, an odd word order, too formal for the scene), it keeps the conversation moving and drops in a short correction underneath, rather than stopping to grade you. That distinction mattered a lot to me: most grammar tools treat every sentence as pass or fail, and that's not how anyone actually learns to talk to people.

Each scenario is pinned to a CEFR level from A1 to C2, so the same "ordering coffee" scene plays out differently depending on where you are — A1 gets short, simple sentences and more patience; C2 gets idioms, faster pacing, and corrections on style rather than just grammar.

Why it's free, and what that actually costs

STEMBridge Speaks runs on Render, and every conversation you have goes through a paid AI model behind the scenes — so "free" here means I'm covering the API and hosting bills myself, not that the underlying service is free to run. There's no ad network and no data-selling plan bolted on to make up the difference; I'd rather keep the scope small and sustainable than take on funding that changes what this is for. If usage grows to where that stops being realistic, I'll be upfront about it on the blog before anything changes.

What's still rough

In the interest of not overselling this: Hebrew's scenario library is noticeably smaller than German's or Spanish's right now, because I built it later and I'm not a native speaker, so I've been slower and more careful with it. Microphone capture is also inconsistent on some older Android browsers — if speech input doesn't pick up cleanly, typing your answer works exactly the same and gets the same correction. And every so often the tutor will over-correct something that was actually fine, usually with very informal or regional phrasing it hasn't seen much of. If that happens to you, the contact page is a real inbox I read.

What's next

None of this is set in stone. The roadmap above is genuinely shaped by what people ask for — if a language or feature isn't listed and you want it, that's useful information for me, not an inconvenience.

If you're using it and something's broken, confusing, or missing — I'd genuinely like to hear about it. See the contact page for how to reach me.