fix: Build errors and tesseract removal

ocr/app/CLAUDE.md

@@ -12,7 +12,7 @@
 
 | Directory | What | When to read |
 | --------- | ---- | ------------ |
-| `engines/` | OCR engine abstraction (PaddleOCR primary, Google Vision fallback, Tesseract compat) | Engine changes, adding new engines |
+| `engines/` | OCR engine abstraction (PaddleOCR primary, Google Vision fallback) | Engine changes, adding new engines |
 | `extractors/` | Data extraction logic | Adding new extraction types |
 | `models/` | Data models and schemas | Request/response types |
 | `patterns/` | Regex and parsing patterns | Pattern matching rules |

ocr/app/config.py

@@ -9,8 +9,6 @@ class Settings:
         self.log_level: str = os.getenv("LOG_LEVEL", "info")
         self.host: str = os.getenv("HOST", "0.0.0.0")
         self.port: int = int(os.getenv("PORT", "8000"))
-        self.tesseract_cmd: str = os.getenv("TESSERACT_CMD", "/usr/bin/tesseract")
-
         # OCR engine configuration
         self.ocr_primary_engine: str = os.getenv("OCR_PRIMARY_ENGINE", "paddleocr")
         self.ocr_confidence_threshold: float = float(

ocr/app/engines/__init__.py

@@ -5,7 +5,6 @@ decoupling extractors from specific OCR libraries.
 
 Engines:
   - PaddleOcrEngine: PaddleOCR PP-OCRv4 (primary, CPU-only)
-  - TesseractEngine: pytesseract wrapper (backward compatibility)
   - CloudEngine: Google Vision TEXT_DETECTION (optional cloud fallback)
   - HybridEngine: Primary + fallback with confidence threshold
 """

ocr/app/engines/base_engine.py

@@ -57,7 +57,7 @@ class OcrEngineResult:
     text: str
     confidence: float  # 0.0-1.0
     word_boxes: list[WordBox]
-    engine_name: str  # "paddleocr", "tesseract", "google_vision"
+    engine_name: str  # "paddleocr", "google_vision"
 
 
 # --- Abstract base ---

ocr/app/engines/engine_factory.py

@@ -11,7 +11,6 @@ logger = logging.getLogger(__name__)
 # Valid engine identifiers (primary engines only; hybrid is constructed separately)
 _ENGINE_REGISTRY: dict[str, str] = {
     "paddleocr": "app.engines.paddle_engine.PaddleOcrEngine",
-    "tesseract": "app.engines.tesseract_engine.TesseractEngine",
     "google_vision": "app.engines.cloud_engine.CloudEngine",
 }
 
@@ -46,7 +45,7 @@ def create_engine(engine_name: str | None = None) -> OcrEngine:
     returns a ``HybridEngine`` that wraps the primary with the fallback.
 
     Args:
-        engine_name: Engine identifier ("paddleocr", "tesseract").
+        engine_name: Engine identifier ("paddleocr", "google_vision").
                      Falls back to ``settings.ocr_primary_engine``.
 
     Returns:

ocr/app/engines/tesseract_engine.py

@@ -1,114 +0,0 @@
-"""Tesseract engine wrapper for backward compatibility."""
-
-import io
-import logging
-
-from app.config import settings
-from app.engines.base_engine import (
-    EngineProcessingError,
-    EngineUnavailableError,
-    OcrConfig,
-    OcrEngine,
-    OcrEngineResult,
-    WordBox,
-)
-
-logger = logging.getLogger(__name__)
-
-
-class TesseractEngine(OcrEngine):
-    """pytesseract wrapper conforming to the OcrEngine interface."""
-
-    def __init__(self) -> None:
-        try:
-            import pytesseract  # type: ignore[import-untyped]
-
-            pytesseract.pytesseract.tesseract_cmd = settings.tesseract_cmd
-            self._pytesseract = pytesseract
-            logger.info("TesseractEngine initialized (cmd=%s)", settings.tesseract_cmd)
-        except ImportError as exc:
-            raise EngineUnavailableError(
-                "pytesseract is not installed. "
-                "Install with: pip install pytesseract"
-            ) from exc
-
-    @property
-    def name(self) -> str:
-        return "tesseract"
-
-    def recognize(self, image_bytes: bytes, config: OcrConfig) -> OcrEngineResult:
-        """Run Tesseract OCR on image bytes."""
-        try:
-            from PIL import Image
-
-            image = Image.open(io.BytesIO(image_bytes))
-
-            # Build Tesseract config string from OcrConfig
-            tess_config = self._build_config(config)
-
-            # Get word-level data
-            ocr_data = self._pytesseract.image_to_data(
-                image,
-                config=tess_config,
-                output_type=self._pytesseract.Output.DICT,
-            )
-
-            word_boxes: list[WordBox] = []
-            texts: list[str] = []
-            confidences: list[float] = []
-
-            for i, text in enumerate(ocr_data["text"]):
-                conf = int(ocr_data["conf"][i])
-                if text.strip() and conf > 0:
-                    normalized_conf = conf / 100.0
-                    word_boxes.append(
-                        WordBox(
-                            text=text.strip(),
-                            confidence=normalized_conf,
-                            x=int(ocr_data["left"][i]),
-                            y=int(ocr_data["top"][i]),
-                            width=int(ocr_data["width"][i]),
-                            height=int(ocr_data["height"][i]),
-                        )
-                    )
-                    texts.append(text.strip())
-                    confidences.append(normalized_conf)
-
-            combined_text = " ".join(texts)
-            avg_confidence = (
-                sum(confidences) / len(confidences) if confidences else 0.0
-            )
-
-            return OcrEngineResult(
-                text=combined_text,
-                confidence=avg_confidence,
-                word_boxes=word_boxes,
-                engine_name=self.name,
-            )
-
-        except (EngineUnavailableError, EngineProcessingError):
-            raise
-        except Exception as exc:
-            raise EngineProcessingError(
-                f"Tesseract recognition failed: {exc}"
-            ) from exc
-
-    def _build_config(self, config: OcrConfig) -> str:
-        """Translate OcrConfig into a Tesseract CLI config string."""
-        parts: list[str] = []
-
-        # Page segmentation mode
-        if config.single_word:
-            parts.append("--psm 8")
-        elif config.single_line:
-            parts.append("--psm 7")
-        else:
-            # Default: assume uniform block of text
-            psm = config.hints.get("psm", 6)
-            parts.append(f"--psm {psm}")
-
-        # Character whitelist
-        if config.char_whitelist:
-            parts.append(f"-c tessedit_char_whitelist={config.char_whitelist}")
-
-        return " ".join(parts)

ocr/app/extractors/manual_extractor.py

@@ -5,9 +5,9 @@ import time
 from dataclasses import dataclass, field
 from typing import Callable, Optional
 
-import pytesseract
 from PIL import Image
 
+from app.engines import create_engine, OcrConfig
 from app.preprocessors.pdf_preprocessor import pdf_preprocessor, PdfInfo
 from app.table_extraction.detector import table_detector, DetectedTable
 from app.table_extraction.parser import table_parser, ParsedScheduleRow
@@ -243,8 +243,9 @@ class ManualExtractor:
 
         # OCR the full page
         try:
-            image = Image.open(io.BytesIO(image_bytes))
-            ocr_text = pytesseract.image_to_string(image)
+            engine = create_engine()
+            ocr_result = engine.recognize(image_bytes, OcrConfig())
+            ocr_text = ocr_result.text
 
             # Mark tables as maintenance if page contains maintenance keywords
             for table in detected_tables:
@@ -358,8 +359,9 @@ class ManualExtractor:
 
             if not text and first_page.image_bytes:
                 # OCR first page
-                image = Image.open(io.BytesIO(first_page.image_bytes))
-                text = pytesseract.image_to_string(image)
+                engine = create_engine()
+                ocr_result = engine.recognize(first_page.image_bytes, OcrConfig())
+                text = ocr_result.text
 
             if text:
                 return self._parse_vehicle_from_text(text)

ocr/app/extractors/vin_extractor.py

@@ -316,8 +316,8 @@ class VinExtractor(BaseExtractor):
             single-line - Treat as a single text line
             single-word - Treat as a single word
 
-        For PaddleOCR, angle classification handles rotated/angled text
-        inherently, replacing the need for Tesseract PSM mode fallbacks.
+        PaddleOCR angle classification handles rotated/angled text
+        inherently, so no PSM mode fallbacks are needed.
 
         Returns:
             List of VIN candidates

ocr/app/preprocessors/vin_preprocessor.py

@@ -93,7 +93,7 @@ class VinPreprocessor:
             gray = cv_image
         steps_applied.append("grayscale")
 
-        # Upscale small images for better OCR (Tesseract needs ~300 DPI)
+        # Upscale small images for better OCR (~300 DPI recommended)
         gray = self._ensure_minimum_resolution(gray)
         steps_applied.append("resolution_check")
 
@@ -129,14 +129,14 @@ class VinPreprocessor:
         )
 
     # Minimum width in pixels for reliable VIN OCR.
-    # A 17-char VIN needs ~30px per character for Tesseract accuracy.
+    # A 17-char VIN needs ~30px per character for reliable OCR accuracy.
     MIN_WIDTH_FOR_VIN = 600
 
     def _ensure_minimum_resolution(self, image: np.ndarray) -> np.ndarray:
         """
         Upscale image if too small for reliable OCR.
 
-        Tesseract works best at ~300 DPI. Mobile photos of VINs may have
+        OCR works best at ~300 DPI. Mobile photos of VINs may have
         the text occupy only a small portion of the frame, resulting in
         low effective resolution for the VIN characters.
         """
@@ -160,7 +160,7 @@ class VinPreprocessor:
         Colored backgrounds have a low min value (e.g. green sticker:
         min(130,230,150) = 130) → inverted to 125 (medium gray).
 
-        The inversion ensures Tesseract always receives dark-text-on-
+        The inversion ensures the OCR engine always receives dark-text-on-
         light-background, which is the polarity it expects.
         """
         b_channel, g_channel, r_channel = cv2.split(bgr_image)
@@ -168,8 +168,8 @@ class VinPreprocessor:
         min_channel = np.minimum(np.minimum(b_channel, g_channel), r_channel)
 
         # Invert so white text (min=255) becomes black (0) and colored
-        # backgrounds (min~130) become lighter gray (~125).  Tesseract
-        # expects dark text on light background.
+        # backgrounds (min~130) become lighter gray (~125).  OCR engines
+        # expect dark text on light background.
         inverted = cv2.bitwise_not(min_channel)
 
         gray = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)

ocr/app/table_extraction/detector.py

@@ -312,7 +312,7 @@ class TableDetector:
         Returns:
             2D list of cell contents
         """
-        # This would use Tesseract on the cropped region
+        # This would use OCR on the cropped region
         # For now, return empty - actual OCR will be done in manual_extractor
         logger.debug(f"Table region: ({table.x}, {table.y}) {table.width}x{table.height}")
         return []

ocr/app/validators/vin_validator.py

@@ -226,7 +226,7 @@ class VinValidator:
         Uses two strategies:
         1. Find continuous 11-20 char alphanumeric runs (handles intact VINs)
         2. Concatenate adjacent short fragments separated by spaces/dashes
-           (handles Tesseract fragmenting VINs into multiple words)
+           (handles OCR fragmenting VINs into multiple words)
 
         Args:
             text: Raw OCR text