ai insight rework #1

ai-insight/extract_text.py

@@ -27,18 +27,25 @@ def extract_docx(file_path):
                 full_text.append(para.text)
         elif tag.endswith('tbl'):
             table = Table(element, doc)
-            table_text = []
+            table_rows = []
             for row in table.rows:
                 row_text = []
                 for cell in row.cells:
-                    text = cell.text.strip()
+                    text = cell.text.strip().replace("|", "/").replace("\n", " ")
                     # Hindari duplikasi text sel gabungan (merged cells) secara berturut-turut
                     if not row_text or row_text[-1] != text:
                         row_text.append(text)
                 if row_text:
-                    table_text.append(" | ".join(row_text))
-            if table_text:
-                full_text.append("\n".join(table_text))
+                    table_rows.append(row_text)
+            if table_rows:
+                # Plan Stage 2: GFM markdown table, header row preserved for RAG.
+                header = table_rows[0]
+                md_lines = [
+                    "| " + " | ".join(header) + " |",
+                    "| " + " | ".join(["---"] * len(header)) + " |",
+                ]
+                md_lines += ["| " + " | ".join(r) + " |" for r in table_rows[1:]]
+                full_text.append("\n" + "\n".join(md_lines) + "\n")
                 
     return "\n\n".join(full_text)