feat(drugs): seed 3,395 drugs from Algerian national nomenclature (Dec 2019)

feat(ocr): Florence-2 as primary handwritten engine — TrOCR as fallback

fix(hf): add paddleocr 2.9.1 — drop opencv pin so paddleocr manages its own opencv-contrib-python

fix(hf): remove paddlepaddle/paddleocr from HF requirements — too many dep conflicts

fix(deps): opencv-python 4.9→4.6.0.66 (paddleocr requires <=4.6.0.66)

fix(deps): paddlepaddle 3.0.0 + paddleocr 2.9.1 — fixes protobuf conflict with onnx

fix(deps): paddlepaddle 2.6.1→2.6.2 (2.6.1 has no Python 3.10 wheel on PyPI)

feat(ocr): PaddleOCR as primary printed engine, Florence-2 as supplement

feat(startup): auto pre-warm Florence-2 after TrOCR loads in background

fix(build): add onnx==1.15.0 — required by optimum.onnxruntime but not pulled in

fix(build): cpu torch + optimum 1.20.0 to fix hard dep conflicts

fix(build): upgrade pip + pin sentencepiece to fix ResolutionImpossible

fix(build): split optimum+onnxruntime, pin onnxruntime==1.17.3, add pip timeout

fix(deps): pin all versions in requirements.txt to stop pip backtracking on HF Space

fix(wake): cut cold-start 180s→90s + better wake widget UX

fix(deps): remove paddlepaddle — protobuf conflict breaks pip install

fix(docker): use python:3.10-slim — paddlepaddle has no Python 3.11 wheel

fix(ocr): pre-warm Florence-2 at startup to prevent cold-start timeout

feat: add Florence-2 VLM to both handwritten and printed OCR pipelines

feat: replace TrOCR-printed with PaddleOCR for printed prescriptions

fix(schema): add trocr-printed to Prescription.method Literal

feat(ocr): add printed prescription mode with TrOCR-printed model

fix(drugs): drop WHO ATC bulk pollution, restore 144 curated Algerian drugs only

feat(ocr): Day 6 — drug cache expanded 62 → 3648 (58×) + matcher false-positive fixes

fix(spaces/Dockerfile): copy schema.py + matcher.py + drugs_cache.json

feat(ocr): Day 3 — /v2 API + DrugMatcher integration + warm-pool ping

fix(ocr-space): cache dir permission + diagnostic error reporting

fix(ocr-space): torch.load weights_only compatibility (PyTorch 2.2+)

fix(ocr-space): async model loading so HF health-check passes

fix(ocr): complete backend overhaul + vision fallback

feat: initial commit — PharMinds Algeria AI pharmacy platform