Fix Docling PP-OCRv6 error + add DB init scripts

- Force TesseractCliOcrOptions for image formats (JPG/PNG/TIFF/BMP)
  to prevent RapidOCR/PP-OCRv6 fallback on docling 2.107
- Add db/init.sql and db/init_docling.sql for database initialization

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

db/init.sql

@@ -0,0 +1,8 @@
+-- markitdown database tables
+CREATE TABLE IF NOT EXISTS conversions (
+    id SERIAL PRIMARY KEY,
+    filename VARCHAR(255) NOT NULL,
+    file_type VARCHAR(50),
+    markdown TEXT,
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);

db/init_docling.sql

@@ -0,0 +1,10 @@
+-- docling database tables
+CREATE TABLE IF NOT EXISTS conversions (
+    id SERIAL PRIMARY KEY,
+    filename VARCHAR(255) NOT NULL,
+    file_type VARCHAR(50),
+    output_format VARCHAR(20) DEFAULT 'markdown',
+    content TEXT,
+    page_count INTEGER,
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);

docling-service/app/services/DoclingService.py

@@ -7,7 +7,7 @@ from fastapi import UploadFile, HTTPException
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy import select
 from docling.document_converter import DocumentConverter, PdfFormatOption
-from docling.datamodel.pipeline_options import PdfPipelineOptions
+from docling.datamodel.pipeline_options import PdfPipelineOptions, TesseractCliOcrOptions
 from docling.datamodel.base_models import InputFormat
 from app.models.ConvertModel import Conversion
 
@@ -21,15 +21,28 @@ CLEANUP_MODEL   = os.getenv("CLEANUP_MODEL", "")
 
 def _build_converter() -> DocumentConverter:
     try:
-        from docling.models.stages.ocr.tesseract_ocr_cli_model import TesseractCliOcrOptions
+        ocr_opts = TesseractCliOcrOptions()
-        pdf_opts = PdfPipelineOptions(do_ocr=True, ocr_options=TesseractCliOcrOptions())
+        pdf_opts = PdfPipelineOptions(do_ocr=True, ocr_options=ocr_opts)
         logger.info("Docling: OCR enabled via Tesseract CLI")
     except Exception as e:
         logger.warning("Docling: Tesseract unavailable (%s) — OCR disabled", e)
+        ocr_opts = None
         pdf_opts = PdfPipelineOptions(do_ocr=False)
-    return DocumentConverter(
+
-        format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pdf_opts)}
+    fmt_options = {InputFormat.PDF: PdfFormatOption(pipeline_options=pdf_opts)}
-    )
+
+    # Force Tesseract for image formats too (prevents RapidOCR/PP-OCRv6 fallback)
+    if ocr_opts is not None:
+        try:
+            from docling.document_converter import ImageFormatOption
+            from docling.datamodel.pipeline_options import ImagePipelineOptions
+            img_opts = ImagePipelineOptions(do_ocr=True, ocr_options=ocr_opts)
+            for fmt in (InputFormat.IMAGE, InputFormat.PNG, InputFormat.JPEG, InputFormat.TIFF, InputFormat.BMP):
+                fmt_options[fmt] = ImageFormatOption(pipeline_options=img_opts)
+        except Exception:
+            pass  # older docling without ImageFormatOption — PDF-only override is sufficient
+
+    return DocumentConverter(format_options=fmt_options)
 
 converter = _build_converter()