Lesingo: Learn Words by Camera
Point your phone at everyday objects and see their English name, pronunciation
and meaning appear over the object, live. Say the word out loud and hear how it
should sound. Object detection runs on the device, so no camera frame leaves
the phone.


What it does
- Point and learn. Detects common objects in the live camera preview and
anchors a compact learning card over each one: the English word, the
translation, an example sentence and the phonetic spelling.
- Speak and compare. Pronounce the word and get feedback; the device
speech service plays the correct pronunciation.
- Build a collection. Every object you find is saved to a collection
organised by room (kitchen, living room, bedroom, bathroom, street,
animals), with levels and a daily streak to keep you practising.
Privacy
Object detection runs locally with a model packaged in the app. Camera images
are processed on the device and are not uploaded. Read the
privacy policy and the
account and data deletion page for the full details.
How it is built
- React Native 0.87 with the New Architecture, one codebase for iOS and Android.
- Feature-first Clean Architecture with MVVM in the presentation layer;
styled-components for the shared theme.
- VisionCamera 5 for the camera session and frame output.
- A local Nitro module (Kotlin on Android) for inference, running MediaPipe
Tasks with EfficientDet-Lite0 int8 across bounded native worker threads.
- Reanimated UI-thread interpolation so overlays follow the objects smoothly;
learning-card metadata is bounded at 30 Hz.
- Device text-to-speech for pronunciation.
On a Samsung SM-M536B (Android 14), four native workers processed 29.8 to 30.2
detector frames per second while the camera preview stayed near 30 FPS. Method
and limitations are in physical-device validation.
Documentation
Lesingo is a side project by Gustavo Rosa.
Source code and issues live on GitHub.