Play ground

Offline rain prediction for 287 Indonesian cities using a pre-trained Random Forest model (40 trees, depth 6). No API calls — runs entirely in-browser.

Per-slot rain prediction (morning/afternoon/evening/night) for 287 Indonesian cities. 4 independent Random Forests with weather persistence feature — fully offline.

Per-slot rain prediction using Gradient Boosted Trees with computed climate features. Improved accuracy over V2 — fully offline.

Per-slot rain prediction using a small Neural Network with learned location embeddings and temporal encoding. First NN experiment — fully offline.

Improved NN experiment: larger network (128→64→32) and embedding (dim=16). Tests whether V3 was capacity-limited — fully offline.

Four independent Neural Networks (one per time slot). Tests whether slot-specific specialization outperforms shared representation — fully offline.

AI-powered 4× image upscaling using Real-ESRGAN x4v3. Runs entirely in-browser via LiteRT.js (WebGPU or WASM). Model ~4 MB + runtime ~9 MB — downloaded once, cached offline.

In-browser image upscaling via Bicubic, Lanczos3, FSR1 (Edge-Adaptive), or EWA Jinc. WebGPU-accelerated — no downloads, no API calls.

Classic 3×3 strategy game. Play PvP (two players on one device) or PvC against a bot with three difficulty levels — Easy, Medium, and unbeatable Hard (Minimax).

Fully offline bill splitting with encrypted storage. Supports equal, custom, and itemized split modes. PIN-protectable.