ASAH

TL;DR — 94%-Accurate Waste Sorting That Runs Offline on Your Phone

ASAH (Aplikasi Sortir Sampah) is an Android app that classifies waste from the camera in real time to guide proper disposal and recycling. A fine-tuned MobileNetV2 model hits 94% accuracy and runs fully on-device via TensorFlow Lite — no internet required — making it usable anywhere.

  Camera frame ──▶ MobileNetV2 (TFLite, on-device) ──▶ waste type + 94% acc.
                         │                                      │
                   no network call                    recycling guidance

The Problem

Most people don't know which bin a given item belongs in, and poor sorting at the source undermines recycling. A useful tool has to give an instant answer, work for non-experts, and keep working in places with weak or no connectivity — ruling out a cloud-only inference approach.

The Approach

• Mobile-first model: MobileNetV2 chosen for its efficiency on phones, fine-tuned on a custom waste-image dataset.
• On-device inference: converted to TensorFlow Lite so classification runs locally — instant and offline.
• Native Android app: built in Kotlin with the TFLite interpreter for smooth real-time performance, plus educational recycling guidance.

Results

• 94% accuracy in waste categorization from a fine-tuned MobileNetV2.
• Real-time, fully offline classification on consumer Android devices.
• Turns correct disposal into a one-tap action for everyday users.