Clairo — AI Voice Dictation for Android
Award-winning team project: an Android app that turns natural speech into clean, formatted text typed into any app — cloud or on-device speech recognition plus LLM cleanup, a floating mic and an AI meeting assistant.
Built by Redoan Uzzaman, Full-Stack Developer & AI Engineer.

Overview
Clairo is an AI writing layer on top of speech-to-text. Instead of dumping a raw transcript, it removes filler words, adds punctuation, matches the tone of the app you are in, and types the result straight into the focused text field — in chat, email, notes or the browser.
How it works
Dictation runs in a native Kotlin foreground service, so it keeps working when the Flutter UI is closed. A trigger (floating mic, Quick Settings tile or notification) starts PCM audio capture; speech is recognised either by a cloud ASR service or on device with SenseVoice via Sherpa-ONNX for private, offline use. An OpenAI-compatible LLM (bring your own key) then formats the transcript for the current context — personal chat, work chat or email — and the Android Accessibility Service inserts it at the cursor, with a clipboard fallback.
The Flutter app (Riverpod, go_router) handles onboarding, writing styles, meetings, history and insights, sharing session data with the native layer through SQLite. Meeting mode records long sessions, transcribes them and offers an assistant that summarises, lists action items and drafts follow-ups.
Recognition
Clairo was built as a team project (original repository: codewith-rafi/clario) and won Third Prize in the Southwest Region of the 2026 China Collegiate Computing Contest — AIGC Innovation Competition.
Key Features
- Third Prize, Southwest Region — 2026 China Collegiate Computing Contest (AIGC Innovation)
- Built as a team — original repository: codewith-rafi/clario
- System-wide dictation: floating mic, Quick Settings tile and notification
- Cloud ASR or private on-device SenseVoice via Sherpa-ONNX
- LLM cleanup with per-app writing styles and tone rewrite
- Bilingual Chinese / English, including mixed speech
- Meeting mode with transcripts and an AI meeting assistant
Tech Stack
- Flutter
- Kotlin
- Sherpa-ONNX
- SenseVoice ASR
- LLM APIs
- Riverpod
- SQLite
- Android Accessibility
Want something like this built?
Redoan Uzzaman takes on AI, automation and full-stack projects for clients worldwide.