update the app

.gitignore

@@ -3,4 +3,5 @@ __pycache__/
 route_venv/
 .env
 /updates
-/github code/
+/github code/
+Project_Inventory_and_Technical_Overview.md

README.md

@@ -16,4 +16,4 @@ short_description: Route image to structured JSON with constraint extraction.
 
 Upload a route document image — get clean structured JSON with constraint classification.
 
-**Model:** `Qwen/Qwen2.5-0.5B-Instruct` (free, open-source, runs on CPU)
+**Model:** `gemini-flash-latest` (via Google GenAI) and `PaddleOCR` (Lightweight)

app.py

@@ -1,15 +1,6 @@
 """
 app.py – OCR Route Data Extraction | Hugging Face Space
 =======================================================
-
-Improved Pipeline:
-1. Engine-specific preprocessing
-2. OCR with EasyOCR OR Lightweight PaddleOCR
-3. OCR normalization
-4. Content-based semantic column classification
-5. Row reconstruction
-6. Rule-based constraint extraction
-7. Structured JSON output
 """
 
 from __future__ import annotations
@@ -23,12 +14,57 @@ import logging
 import re
 import time
 from statistics import median
-from typing import Optional
+from typing import List, Optional
 
 import cv2
 import gradio as gr
 import numpy as np
 from PIL import Image
+from pydantic import BaseModel, ConfigDict, Field, ValidationError
+
+
+class ConstraintModel(BaseModel):
+    model_config = ConfigDict(populate_by_name=True)
+
+    type: str
+    action: str
+    from_: str = Field(alias="from")
+    to: str
+    priority: Optional[str] = None
+
+
+class StepModel(BaseModel):
+    step: int = Field(ge=1)
+    given_miles: float = Field(ge=0)
+    road: str
+    instruction: str
+    distance: float = Field(ge=0)
+    est_time: str = Field(pattern=r"^\d{2}:\d{2}$")
+    constraints: List[ConstraintModel] = Field(default_factory=list)
+
+
+class AccuracyMetricsModel(BaseModel):
+    ocr_confidence: Optional[float] = Field(default=None, ge=0, le=100)
+    extraction_score: Optional[float] = Field(default=None, ge=0, le=100)
+    total_accuracy: Optional[float] = Field(default=None, ge=0, le=100)
+    gemini_confidence: Optional[str] = None
+
+
+class RouteExtractionModel(BaseModel):
+    source: str
+    extracted_at: str
+    ocr_engine: str
+    extraction: str
+    accuracy_metrics: Optional[AccuracyMetricsModel] = None
+    total_steps: int = Field(ge=0)
+    total_miles: float = Field(ge=0)
+    total_time: str = Field(pattern=r"^\d{2}:\d{2}$")
+    steps: List[StepModel]
+    warning: Optional[str] = None
+    gemini_error: Optional[str] = None
+
+
+ROUTE_EXTRACTION_JSON_SCHEMA = RouteExtractionModel.model_json_schema()
 
 
 # ============================================================
@@ -753,6 +789,7 @@ def run_gemini_correction(image_pil: Image.Image, initial_json: dict, api_key: s
         
     client = genai.Client(api_key=api_key)
     
+    schema_text = json.dumps(ROUTE_EXTRACTION_JSON_SCHEMA, indent=2)
     prompt = f"""
 You are an expert OCR data extraction and validation system for route documents.
 You are given an image of a route document and an initial JSON extraction. The basic OCR system is flawed and often misses data or misinterprets columns.
@@ -770,6 +807,9 @@ CRITICAL EXTRACTION RULES:
 
 Output ONLY valid JSON matching the exact schema.
 
+JSON SCHEMA (MUST MATCH):
+{schema_text}
+
 Initial JSON (FLAWED - USE FOR REFERENCE ONLY):
 {json.dumps(initial_json, indent=2)}
 """
@@ -779,11 +819,17 @@ Initial JSON (FLAWED - USE FOR REFERENCE ONLY):
         contents=[image_pil, prompt],
         config=types.GenerateContentConfig(
             response_mime_type="application/json",
+            response_schema=RouteExtractionModel,
             temperature=0.0
         )
     )
     
     try:
+        parsed = getattr(response, "parsed", None)
+        if parsed is not None:
+            if hasattr(parsed, "model_dump"):
+                return parsed.model_dump(by_alias=True)
+            return parsed
         return json.loads(response.text)
     except json.JSONDecodeError:
         raise ValueError(f"Failed to parse JSON from Gemini response:\n{response.text}")
@@ -986,6 +1032,11 @@ def run_pipeline(
         "steps": steps
     }
 
+    try:
+        result = RouteExtractionModel.model_validate(result).model_dump(by_alias=True)
+    except ValidationError as e:
+        log.warning("Schema validation failed for rule-based output: %s", e)
+
     if "Gemini" in ocr_engine or api_key.strip():
         progress(0.85, desc="Validating the JSON output ...")
         try:
@@ -999,6 +1050,7 @@ def run_pipeline(
                 result["accuracy_metrics"] = {}
             result["accuracy_metrics"]["total_accuracy"] = 98.5
             result["accuracy_metrics"]["gemini_confidence"] = "High"
+            result = RouteExtractionModel.model_validate(result).model_dump(by_alias=True)
         except Exception as e:
             import traceback; traceback.print_exc()
             error_text = str(e)