lesingo

Roadmap

The roadmap favors small, reviewable milestones that each produce visible technical evidence. Dates are intentionally omitted until the foundation is validated on a physical Android device.

Milestone 0: Repository foundation

Status: complete

• initialize the local Git repository;
• document the product intent;
• define feature-first Clean Architecture boundaries;
• record the initial Android camera and ML decisions;
• establish semantic commits and contribution guidance.

Evidence: documentation, ADRs, repository structure, and atomic history.

Milestone 1: Android React Native baseline

Status: complete

• create the React Native Android application with the New Architecture;
• pin Node, Java, Gradle, Android SDK, React Native, and package-manager versions;
• add linting, formatting, type checking, and unit-test commands;
• create a minimal application shell through the app composition root;
• verify a debug build on a physical Android device.

Acceptance criteria:

• a clean checkout can follow documented setup steps;
• all quality commands run locally;
• the Android debug app launches without camera or ML dependencies.

Evidence: React Native 0.87 app with the New Architecture and Hermes, automated quality gate, successful debug build, installation, launch, process check, UI hierarchy check, and screenshot from a physical Android 14 device. See the validation record.

Milestone 2: Camera pipeline

Status: in progress

• integrate VisionCamera 5 and camera permission handling;
• render the back-camera preview;
• add Camera and Settings navigation with camera lifecycle control;
• structure presentation with MVVM and a shared styled-components theme;
• validate a continuous five-minute preview session;
• configure a low-resolution FrameOutput independently from the preview;
• drop busy frames and dispose them on success and failure paths;
• document frame disposal and orientation mapping behavior.

Acceptance criteria:

• preview remains responsive for a five-minute session;
• busy processing drops frames rather than creating a queue;
• no raw pixel buffer reaches the JavaScript thread.

Milestone 3: Native object detection

Status: in progress

• scaffold the Android Nitro detector module;
• record the selected model source, version, license, and checksum;
• integrate MediaPipe Tasks and EfficientDet-Lite0 int8;
• map native results to the typed detector contract;
• measure an initial inference sample on physical hardware;
• record latency percentiles and a known-object physical test set.

Acceptance criteria:

• common test objects produce labels, scores, and bounding boxes;
• model resources close correctly across reload and camera deactivation;
• latency percentiles and device details are documented.

Milestone 4: Stable interactive overlay

Status: in progress

• convert frame coordinates into preview coordinates;
• assign stable track identities using label and bounding-box overlap;
• smooth visual position and size on the UI thread;
• render accessible live vocabulary cards;
• handle stale detections deterministically.

Acceptance criteria:

• overlays align through orientation and preview crop scenarios;
• labels remain visually stable during ordinary hand movement;
• each visible target shows the expected word, meaning, pronunciation, and confidence without interaction.

Milestone 5: Learning experience

Status: in progress

• define the first curated vocabulary catalog for common detector labels;
• show word, pronunciation, meaning, and confidence in the camera layer;
• integrate device text-to-speech;
• add scanning, unknown-label, permission, and failure states;
• polish motion, haptics, and visual hierarchy.

Acceptance criteria:

• the critical journey works offline after installation;
• every supported visible label has reviewed learning content;
• pronunciation and details remain usable with the camera paused.

Milestone 6: Performance engineering evidence

Status: planned

• add a repeatable release benchmark and structured metric capture;
• record detector latency percentiles and end-to-end result freshness;
• measure native allocation pressure and bitmap reuse;
• compare worker counts and detector resolutions against CPU, memory, battery, and thermal cost;
• profile React commits and verify UI-thread interpolation behavior;
• validate ten- and thirty-minute physical-device sessions;
• measure cold startup, release size, R8 impact, and 16 KB alignment.

Acceptance criteria:

• before-and-after runs use the same device, scene, build type, and protocol;
• published numbers include distributions and resource costs, not only FPS;
• the chosen configuration is justified by an efficiency curve;
• raw evidence, failed experiments, and limitations are reproducible;
• claims satisfy the performance publication gate.

Milestone 7: Public technical release

Status: planned

• select and add an open-source license;
• add automated Android and JavaScript quality checks;
• test setup from a clean clone;
• publish performance measurements and known limitations;
• record a concise demo and prepare the technical write-up;
• create the public GitHub repository using the intended personal account.

Acceptance criteria:

• repository history is understandable commit by commit;
• no secret, signing material, or personal local configuration is tracked;
• README explains architecture, setup, trade-offs, and evidence;
• release artifacts link back to reproducible source.