Add NoneType safety checks in _confidence_badge to prevent extraction card formatting crash

Add opencv-python-headless requirement for cloud VLM grounding layout processing

Replace VLM grounding with OpenCV line layout analysis and filter non-drug crops using is_valid_drug_line

Fix bounding box grounding coordinate detection, crop OCR line parser, and render all extracted medicines in UI

Implement segment-and-crop OCR with green bounding box drawing and display

Implement multi-drug extraction, validation, and dropdown selection UI

Refactor medicine search with comprehensive NAME_TO_MED lookup, fix legibility, add openai requirement, add new brand mappings

Multi-pass OCR + Nemotron 8B: drug-focused pass, Llama-3.1-Nemotron-Nano-8B-v1

Fix OCR prompt: prioritize drug names over Bengali headers, multilingual prescription handling

Full structured prescription extraction: two-pass OCR (MiniCPM-V + Nemotron), HIPAA JSON schema, controlled substance detection, compliance banners

Update code to fresh version from parent project

UI/UX Overhaul: Premium cyber-medical dark theme, glassmorphic styles, custom SVG metric icons, active pipeline stepper, and Plotly gauge sync

Use HF Nemotron Mini for local validation

Fallback Nemotron validation to NVIDIA NIM

Install Mamba runtime for Nemotron

Add local Nemotron validation UI

Require review for low confidence medicine matches

Use live retrieval result instead of Neuroxen fallback

Run MiniCPM OCR on Hugging Face GPU

Center KPI rail on Hugging Face

Polish first screen layout and KPI rail

Apply custom styling in Space entrypoint

Add ZeroGPU startup compatibility

Fix Gradio 6 image and launch compatibility

Deploy PharmaCopilot Gradio workflow with bucket assets

initial commit