Free Mobile Receipt Scanner Runs Fine-Tuned 0.8B Vision Model On-Device
An engineer has developed a free, privacy-focused Android application that utilizes a fine-tuned 0.8B vision model to extract detailed line items from grocery receipts into structured JSON. The app runs the model completely locally on the mobile device, with an iOS version pending review. This project demonstrates the growing practicality of Edge AI, allowing users to perform complex document parsing and data extraction locally without compromising personal financial privacy or relying on cloud APIs. It highlights how small language and vision models are becoming lightweight enough for everyday consumer hardware. The application uses fine-tuned versions of Qwen3.5 0.8B and OvisOCR2 specifically trained on receipt data to parse store names, dates, totals, and individual line items. Additionally, it offers an option to connect to a user's own OpenAI API-compatible server if the local processing proves too memory-intensive.
## BACKGROUND
Edge AI refers to running machine learning models directly on local hardware devices like smartphones, which eliminates cloud latency and protects user privacy by processing data offline. Vision-language models combine image understanding with text generation, enabling tools like OCR and document parsers to read unstructured layouts such as receipts and convert them into structured data formats.