AI-powered appliance repair assistant, built entirely on-device for the Gemma 4 Good Hackathon.
FixGemma is a Flutter application that leverages Gemma 4 (e2b & e4b models) via the Cactus AI inference engine to provide users with private, completely offline, multimodal repair assistance. By empowering people to repair their own appliances, FixGemma promotes sustainability, fights e-waste, and champions the "Right to Repair".
- Fully On-Device Inference: No internet required for AI generation. 100% private.
- Multimodal Inputs: Describe the issue via text, record your voice, or take a picture of the broken appliance.
- Structured Output: Generates step-by-step repair guides (safety warnings, tools required, steps, and tips) by streaming and parsing JSON in real-time.
- Background Downloads: Robust management of large model weight downloads, ensuring completion even when the app is minimized.
- Beautiful UI: Modern, responsive UI with frosted glass effects and card-based step-by-step navigation.
- Framework: Flutter (Riverpod for state management, Hive for local storage)
- AI Engine: Cactus AI inference engine
- Models:
fixgemma4-e4b-int4(6.9 GB) - Finetuned version of gemma 4 e4b.fixgemma4-e2b-int4(4.0 GB) - Finetuned version of gemma 4 e2b.
- Flutter SDK (>=3.3.0)
- Android Studio / Xcode for device emulation
- A physical device (recommended) with sufficient RAM and storage for on-device inference.
- Clone the repository:
git clone https://github.com/sanjayram-a/fixgemma.git cd fixgemma - Install dependencies:
flutter pub get
- Run the app:
flutter run
- Upon first launch, navigate to settings to download your preferred Gemma 4 model or from home screen.
For a detailed overview of our Hackathon submission, problem statement, and technical challenges, please read the Hackathon Writeup.