Deploy to Hugging Face Spaces (Docker) + cut OCR memory (200 DPI, close doc)

.dockerignore

@@ -0,0 +1,9 @@
+.venv/
+__pycache__/
+*.pyc
+.pytest_cache/
+.git/
+.env
+.env.local
+_work/
+tests/

Dockerfile

@@ -0,0 +1,30 @@
+# Extractor service for Hugging Face Spaces (Docker SDK).
+# HF free CPU tier gives 16GB RAM — no more OOM from PDF/OCR rendering.
+FROM python:3.13-slim
+
+# System dep: tesseract for OCR (the engine shells out to the `tesseract` CLI).
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends tesseract-ocr \
+    && rm -rf /var/lib/apt/lists/*
+
+# HF Spaces run containers as uid 1000; set up a matching non-root user + writable home.
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:$PATH \
+    PYTHONUNBUFFERED=1 \
+    EBS_WORK_DIR=/home/user/app/_work
+
+WORKDIR /home/user/app
+
+# Install Python deps first (better layer caching).
+COPY --chown=user:user requirements.txt .
+RUN pip install --no-cache-dir --user -r requirements.txt
+
+# App code + a writable work dir for uploads/OCR temp images.
+COPY --chown=user:user . .
+RUN mkdir -p /home/user/app/_work
+
+# HF Spaces expect the app on port 7860.
+EXPOSE 7860
+CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -0,0 +1,16 @@
+---
+title: Export Bank Statement Extractor
+emoji: 📄
+colorFrom: green
+colorTo: blue
+sdk: docker
+app_port: 7860
+pinned: false
+---
+
+# Export Bank Statement — Extractor
+
+FastAPI service: PDF bank statement in → reconciled statement data out (JSON / .xlsx).
+Stateless; auth/quota/billing are owned by the web app. OCR via the `tesseract` CLI.
+
+**Endpoints:** `GET /health` · `POST /convert` (header `x-api-key: $EBS_API_KEY`)

app/engine/converter.py

@@ -177,10 +177,14 @@ class BankStatementConverter:
         import fitz
 
         doc = fitz.open(self.pdf_path)
-        pix = doc[page_index].get_pixmap(dpi=300)
-        # write the temp image next to the PDF (tesseract/leptonica can't read /tmp here)
-        png = self.pdf_path.parent / f".ocr_p{page_index}.png"
-        png.write_bytes(pix.tobytes("png"))
+        try:
+            # 200 DPI is enough for OCR; 300 rendered ~30MB/page and spiked RAM -> OOM on the host.
+            pix = doc[page_index].get_pixmap(dpi=200)
+            # write the temp image next to the PDF (tesseract/leptonica can't read /tmp here)
+            png = self.pdf_path.parent / f".ocr_p{page_index}.png"
+            png.write_bytes(pix.tobytes("png"))
+        finally:
+            doc.close()  # release the page/pixmap immediately so RAM doesn't grow page-by-page
         try:
             proc = subprocess.run(["tesseract", str(png), "stdout", "tsv"],
                                   stdout=subprocess.PIPE, stderr=subprocess.DEVNULL)