Dev Fine-Tunes 0.8B Vision Models for Free On-Device Receipt Scanning
Trick_Ad_1135 · reddit · 2026-10-12
To track grocery spending, a developer built an Android receipt-scanning app that runs entirely on-device with no data uploaded:
- Models: Fine-tuned Qwen3.5 0.8B and OvisOCR2 (itself a fine-tune of Qwen3.5 0.8B) to read receipt photos into JSON — store, date, total, and each line item.
- Differentiation: Most expense apps only offer coarse categories like "food"; this one drills into item-level detail, e.g. 150 欧 spent on coffee this month.
- Open-source plans: The app is free, and the model will be open-sourced for self-hosted servers; it also supports connecting your own OpenAI-compatible API as a fallback when compute runs short. The author invites users to donate receipt photos to improve the model.
A textbook indie case of "small model + edge deployment + niche use case"; an iOS version is awaiting Apple review.
More from Apps
- Compressing a ~500MB RoBERTa AI-text detector to ~97MB that runs fully in-browser — Alternative_Detail31 · 2026-10-12
- PwC upgrades paper PDF reader with faster loading and chapter jump navigation — NielsRogge · 2026-10-12
- From Poke to OpenAI's Dots: the personal messaging agent era is here — sujingshen · 2026-10-12
- Reddit Users Push Back on Gemini Requiring Google Workspace for Simple Writing — Any_Professional3884 · 2026-10-12
- 7 Free GitHub Repos That Replace Paid Apps, From Voice Cloning to Local LLMs — alex_verem · 2026-10-12
- Voyager: an open agent harness for creative work that reuses your model subscriptions — ravisparikh · 2026-10-12