TL;DR

I built FreeSay — an AI-powered speaking practice app for language learners. Real-time conversation with an AI teacher, 43 UI languages, 15 learning languages, $2.99/month (about 50x cheaper than Cambly's $150/month plan).

Live: https://www.freesay.app

Why I built it

Cambly, italki, Preply — all great apps, but priced for developed-country wallets. A student in India, Indonesia, or Pakistan who wants to practice English speaking usually has two options: expensive human tutors or silent flashcard apps. No middle ground for actually talking.

With LLMs hitting "good enough" quality for conversational language practice, the price floor collapsed. I set out to build what a $3/month AI tutor looks like.

Stack & decisions

• Frontend: React (web) + native iOS/Android
• Backend: Node + Postgres, streaming voice pipeline
• LLM: a mix routed by cost/latency; speech-to-text + TTS swapped per region
• Regional pricing baked in from day 1 — 250 INR in India, 45,000 IDR in Indonesia, 800 PKR in Pakistan. Flat $2.99 doesn't work when median income varies 20x.
• Low-end phone first: works on 2GB RAM Android, degraded-network fallbacks

What surprised me

1. Voice latency matters more than LLM quality. Users forgive a dumber tutor; they don't forgive a 4-second pause.
2. "Beginner English" is a huge underserved segment. Most AI language apps implicitly target people who already speak English well enough to self-direct.
3. Regional pricing isn't just "charge less" — it changes churn, payment method distribution, and even what topics users ask about.

What I'd love feedback on

• Pricing: is $2.99 sustainable or am I racing to the bottom?
• Retention patterns in emerging markets — what's worked for others?
• Anyone else shipping AI products with per-country pricing? How are you handling it?

Try it: freesay.app — would love any feedback.