Add segment_special_appeal script

Author: edubezerra

.vscode/settings.json

@@ -0,0 +1,6 @@
+{
+    "python-envs.defaultEnvManager": "ms-python.python:conda",
+    "python-envs.defaultPackageManager": "ms-python.python:conda",
+    "python-envs.pythonProjects": [],
+    "python.analysis.typeCheckingMode": "basic"
+}

src/segment_special_appeal.py

@@ -0,0 +1,91 @@
+import json
+from openai import OpenAI
+
+
+def segment_special_appeal(
+    appeal_text: str,
+    model: str = "gpt-4.1-mini",
+    temperature: float = 0.0
+) -> dict:
+    """
+    Segments a Brazilian special appeal into rhetorical/legal sections.
+
+    Parameters
+    ----------
+    appeal_text : str
+        Full text of the special appeal.
+    model : str
+        OpenAI model to use.
+    temperature : float
+        Sampling temperature (keep at 0 for determinism).
+
+    Returns
+    -------
+    dict
+        JSON-like dict with segmented sections.
+    """
+
+    client = OpenAI()
+
+    system_prompt = (
+        "You are a legal text analysis assistant. "
+        "Your task is to segment Brazilian legal documents based on their rhetorical and functional structure. "
+        "Do NOT summarize, classify, interpret legal outcomes, or infer themes. "
+        "Only segment the text into structurally meaningful sections."
+    )
+
+    user_prompt = f"""
+Segment the following Brazilian special appeal into the sections listed below.
+
+Sections:
+- identification
+- procedural_history
+- facts
+- legal_issues
+- legal_arguments
+- requests
+- other
+
+Instructions:
+- Preserve the original text verbatim in each section.
+- Do NOT rewrite, paraphrase, or summarize.
+- If a section is not present, return it as an empty string.
+- Return the output strictly as a valid JSON object.
+- Do not include any explanation or commentary.
+
+TEXT:
+\"\"\"
+{appeal_text}
+\"\"\"
+"""
+
+    response = client.chat.completions.create(
+        model=model,
+        temperature=temperature,
+        messages=[
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt},
+        ]
+    )
+
+    raw_output = response.choices[0].message.content
+
+    try:
+        segmented = json.loads(raw_output)
+    except json.JSONDecodeError as e:
+        raise ValueError("Model output is not valid JSON") from e
+
+    return segmented
+
+
+if __name__ == "__main__":
+    # Example usage (for testing)
+    with open("recurso_especial.txt", "r", encoding="utf-8") as f:
+        text = f.read()
+
+    result = segment_special_appeal(text)
+
+    with open("recurso_segmentado.json", "w", encoding="utf-8") as f:
+        json.dump(result, f, ensure_ascii=False, indent=2)
+
+    print("Segmentation completed. Output saved to recurso_segmentado.json")