Fix: serialize structured_data and trace as JSON strings in dataset traces

Add OCR: image + scanned PDF support via tesseract + pdf2image

Fix: flexible medication/allergy extraction for real-world document formats

Fix: auto-load sample docs at startup so state survives restarts

Fix: add missing import re to app.py (caused NameError in safety_check)

Fix: regex extraction on original text + noise filter

Fix: strip non-ASCII chars before sending to llama-server

Fix: targeted retrieval for safety context + sharper extraction prompt

Fix: remove --flash-attn flag, increase timeout to 300s

Fix: use gradio 6.0.0 sdk_version - resolves HfFolder error

Fix: upgrade to gradio 5.x, drop conflicting pins

Upload app.py