Spatial sorting

src/filler.py

@@ -1,16 +1,58 @@
-from pdfrw import PdfReader, PdfWriter
+from pdfrw import PdfReader, PdfWriter, PdfName
 from src.llm import LLM
 from datetime import datetime
 
-
 class Filler:
     def __init__(self):
         pass
 
+    def sort_annots_by_cluster(self, annots, y_tolerance=10.0):
+        """
+        Groups annotations into 'rows' based on their Y-coordinate (within a tolerance).
+        Then sorts each row by the X-coordinate (left-to-right).
+        """
+        # First, roughly sort them top-to-bottom
+        rough_sort = sorted(annots, key=lambda a: -float(a.Rect[1]))
+
+        rows = []
+        current_row = []
+        current_y = None
+
+        for annot in rough_sort:
+            y_val = -float(annot.Rect[1])
+
+            # If it's the first item, start the row
+            if current_y is None:
+                current_y = y_val
+                current_row.append(annot)
+                continue
+
+            # If the current item is within the Y-tolerance of the current row
+            if abs(y_val - current_y) <= y_tolerance:
+                current_row.append(annot)
+            else:
+                # We've moved to a new row. Save the current one and start fresh.
+                rows.append(current_row)
+                current_row = [annot]
+                current_y = y_val
+
+        # Don't forget the last row
+        if current_row:
+            rows.append(current_row)
+
+        # Now, sort each row left-to-right (by X coordinate)
+        final_sorted = []
+        for row in rows:
+            sorted_row = sorted(row, key=lambda a: float(a.Rect[0]))
+            final_sorted.extend(sorted_row)
+
+        return final_sorted
+
+
     def fill_form(self, pdf_form: str, llm: LLM):
         """
         Fill a PDF form with values from user_input using LLM.
-        Fields are filled in the visual order (top-to-bottom, left-to-right).
+        Fields are filled in visual order using clustered spatial sorting.
         """
         output_pdf = (
             pdf_form[:-4]
@@ -19,9 +61,9 @@ def fill_form(self, pdf_form: str, llm: LLM):
             + "_filled.pdf"
         )
 
-        # Generate dictionary of answers from your original function
+        # Generate dictionary of answers
         t2j = llm.main_loop()
-        textbox_answers = t2j.get_data()  # This is a dictionary
+        textbox_answers = t2j.get_data() 
 
         answers_list = list(textbox_answers.values())
 
@@ -31,22 +73,33 @@ def fill_form(self, pdf_form: str, llm: LLM):
         # Loop through pages
         for page in pdf.pages:
             if page.Annots:
-                sorted_annots = sorted(
-                    page.Annots, key=lambda a: (-float(a.Rect[1]), float(a.Rect[0]))
-                )
+                # -- NEW CLUSTER SORTING LOGIC --
+                sorted_annots = self.sort_annots_by_cluster(page.Annots)
 
                 i = 0
                 for annot in sorted_annots:
                     if annot.Subtype == "/Widget" and annot.T:
                         if i < len(answers_list):
-                            annot.V = f"{answers_list[i]}"
-                            annot.AP = None
+                            val = str(answers_list[i])
+
+                            # CHECKBOX LOGIC (From previous PR)
+                            if annot.FT == "/Btn":
+                                if val.lower() in ["yes", "on", "true"]:
+                                    annot.V = PdfName("Yes") 
+                                    annot.AS = PdfName("Yes")
+                                else:
+                                    annot.V = PdfName("Off")
+                                    annot.AS = PdfName("Off")
+
+                            # STANDARD TEXT LOGIC
+                            else:
+                                annot.V = f"{val}"
+                                annot.AP = None 
+
                             i += 1
                         else:
-                            # Stop if we run out of answers
                             break
 
         PdfWriter().write(output_pdf, pdf)
 
-        # Your main.py expects this function to return the path
-        return output_pdf
+        return output_pdf

src/llm.py

@@ -46,7 +46,7 @@ def build_prompt(self, current_field):
 
     def main_loop(self):
         # self.type_check_all()
-        for field in self._target_fields.keys():
+        for field in self._target_fields:
             prompt = self.build_prompt(field)
             # print(prompt)
             # ollama_url = "http://localhost:11434/api/generate"

src/main.py

@@ -3,6 +3,7 @@
 from commonforms import prepare_form 
 from pypdf import PdfReader
 from controller import Controller
+from typing import Union
 
 def input_fields(num_fields: int):
     fields = []
@@ -68,7 +69,7 @@ def run_pdf_fill_process(user_input: str, definitions: list, pdf_form_path: Unio
 if __name__ == "__main__":
     file = "./src/inputs/file.pdf"
     user_input = "Hi. The employee's name is John Doe. His job title is managing director. His department supervisor is Jane Doe. His phone number is 123456. His email is jdoe@ucsc.edu. The signature is <Mamañema>, and the date is 01/02/2005"
-    fields = ["Employee's name", "Employee's job title", "Employee's department supervisor", "Employee's phone number", "Employee's email", "Signature", "Date"]
+    descriptive_fields = ["Employee's name", "Employee's job title", "Employee's department supervisor", "Employee's phone number", "Employee's email", "Signature", "Date"]
     prepared_pdf = "temp_outfile.pdf"
     prepare_form(file, prepared_pdf)
 
@@ -80,4 +81,4 @@ def run_pdf_fill_process(user_input: str, definitions: list, pdf_form_path: Unio
         num_fields = 0
 
     controller = Controller()
-    controller.fill_form(user_input, fields, file)
+    controller.fill_form(user_input, descriptive_fields, file)