English Circle
An AI English-conversation tutor for Mexican Spanish speakers. Live group voice classes with “Alex,” an AI teacher who understands Mexican slang, corrects you gently in real time, and sends a personal report after every class.
Alex is running right now. Talk to him for a minute in your browser — no sign-up, no install.
▶ Launch the demoThe problem
Millions of Mexicans want to speak English, but the options fail at the one thing that matters — real conversation. Apps gamify vocabulary but never make you talk; private tutors cost a fortune; and generic chatbots have no structure and no cultural awareness (they flag Mexican slang as “errors”).
What I built
A conversational teacher you actually talk to. Alex listens, replies in natural spoken English in well under a second, and gently steers the conversation — while quietly catching the things worth correcting and weaving them back in without breaking your flow:
- Real-time voice — replies in well under a second (~500 ms), so it feels like talking to a person, not waiting on a chatbot.
- Mexican-Spanish aware — slang like chido or no manches is welcomed and bridged to English, never flagged as an error.
- Gentle, in-flow corrections — mistakes are noticed and brought back into the conversation, never thrown in your face.
- Per-session reports — every class ends with a personal report: corrections, expressions used, and progress.
- Group classes — LiveKit voice rooms with breakout pairs, so everyone actually speaks.
It speaks the way the city speaks
Generic tutors flag Mexican slang as a mistake. Alex hears it the way a chilango would — and bridges it into natural English without breaking your flow.
“¿Qué onda, profe?”
Alex keeps the class flowing in natural English instead of freezing on words a textbook never taught.
Engineering highlights
- Sub-second spoken responses while still tracking and correcting in the background — the hard part of a voice tutor that feels live.
- A cost/latency optimization that cut per-turn model calls ~80% after profiling production logs — keeps the live demo cheap enough to leave open.
- Two interchangeable voice paths, built and benchmarked against each other — a speech-to-speech path on Gemini Live native audio, and a composed pipeline (Deepgram STT → Gemini 2.5 Flash → ElevenLabs TTS). Selectable per deployment, so the latency/quality/cost tradeoff is a configuration decision rather than a rewrite.
- Real-time voice transport on LiveKit Agents; deployed as a single-image API + worker service on Railway, with a Next.js front end on Vercel.
- 600+ automated tests across the conversation and voice layers.