Ticha AI

Topic selection page
Session screen
Parent's dashboard
Landing page
Quiz screen
Child's dashboard

Motivation

Swahili is spoken by more than 200 million people in East Africa, but most language-learning apps require kids to read and type, skills that millions of African kids are still learning. In the same way that children learn their first language, I wanted to create something that would allow them to just talk and learn. Ticha, which means "teacher" in Swahili, is that instrument.

What it achieves

Ticha is a live AI voice tutor that teaches Swahili and English to African kids. When a child presses a button, Ticha calls them by name, teaches five vocabulary words through interactive questions and vivid stories, plays a review game, and then administers a picture quiz all by voice. Not a word. Do not type. Simply speak.

A dashboard is provided to parents so they can monitor each session, words learned, XP earned, and streak.

How I built it

Gemini 2.5 Flash Live (gemini-2.5-flash-native-audio-latest) uses the Google GenAI SDK to stream voice in both directions in real time. Railway has deployed Next.js 15 (App Router, TypeScript). Supabase for database (child profiles, session history, XP) and authentication Web Audio API – dual AudioContext configuration: gapless PCM16 playback at 24 kHz, microphone capture at 16 kHz via AudioWorkletA 5-step CEFR A1 lesson flow with age-adaptive pacing, pronunciation guides, and language-lock rules is enforced by a ~12KB structured system prompt. Challenges I faced

Scheduling is necessary for gapless audio playback, which involves streaming PCM16 segments without silence intervals. Web Audio clock-based AudioBufferSourceNode chainsBarge-in: leaving the microphone open so kids can naturally cut Ticha off in the middle of her sentence Language consistency: The prompt needed a strict language-lock system to prevent Gemini from switching languages mid-lesson. The phrase "Tutaonana" (goodbye in Swahili) is used as a dependable session-end signal in auto-end detection.

What I discovered in this process

In contrast to text AI, real-time voice AI primarily focuses on latency, audio quality, and natural conversational flow. Another layer is added when building for children: each response needs to be designed with warmth, encouragement, and patience.

What comes next, Ticha AI?

Additional African languages: Zulu, Twi, and Yoruba-Offline mode in places with poor connectivity. A dashboard for teachers in African classrooms.