Know Your Sh#t
Restaurant menus are a pain to digitize. Most owners retype everything by hand or pay someone to do it. The goal: upload a PDF menu and get a structured, editable digital menu back — automatically.
Stack
Approach
Built a processing pipeline: PDF upload → page-by-page OCR with Tesseract → text analysis via a local LLM (Ollama/Mistral) → auto-generated menu sections and items. Real-time progress updates stream to the browser as each page processes.
Results
Upload a PDF, watch it process in real-time, review the extracted menu, approve or edit, done. No manual data entry. The entire pipeline runs locally — no external API costs.
Under the Hood
Rails 8 with Hotwire (Turbo Streams) for real-time processing updates without JavaScript. Tesseract OCR extracts text from each PDF page, then Ollama runs Mistral locally to parse unstructured text into JSON menu structure. Active Job + Solid Queue for async processing. Confidence scoring on OCR results flags pages that need manual review.