Visionia OCR
An OCR application featuring a custom-trained model for text recognition, with AI-powered chat functionality using locally hosted Ollama models to interact with extracted content. Includes camera capture for handwriting-to-text conversion. Built with Expo/React Native, Firebase, Flask API, Ollama, and Cloudflare Tunnel.
Computer Vision
AI Chat
Mobile App
Problem
Users need to digitize handwritten notes and printed documents, but existing OCR solutions lack the ability to immediately interact with and query the extracted content. Additionally, privacy concerns arise when sensitive documents are processed through cloud-based services.
Solution
Built a mobile OCR app with three custom-trained recognition models (TrOCR English, TrOCR fine-tuned French, and CRNN) for accurate text extraction. Integrated four self-hosted AI models—Llama 3.2:1b, Gemma 2:4b, DeepSeek-R1, and Mistral—via Ollama for intelligent document analysis. Users capture images via camera, extract text with custom models, and immediately chat with AI about the content, all processed through self-hosted Flask API and Cloudflare Tunnel without sending data to external servers.
