llm_client.py

"""
Azure OpenAI client wrapper for LLM-based analysis.
"""

import os
import json
from openai import AzureOpenAI
from dotenv import load_dotenv
from prompts import FOOTBALL_ANALYSIS_SYSTEM_PROMPT, FOOTBALL_ANALYSIS_USER_PROMPT

load_dotenv()

AZURE_OPENAI_KEY = os.getenv("AZURE_OPENAI_KEY")
AZURE_OPENAI_ENDPOINT = os.getenv("AZURE_OPENAI_ENDPOINT")
AZURE_OPENAI_API_VERSION = os.getenv("AZURE_OPENAI_API_VERSION", "2024-12-01-preview")
AZURE_OPENAI_DEPLOYMENT = os.getenv("AZURE_OPENAI_DEPLOYMENT", "gpt-5.2-chat")


def get_openai_client():
    """Create and return an Azure OpenAI client."""
    return AzureOpenAI(
        api_key=AZURE_OPENAI_KEY,
        azure_endpoint=AZURE_OPENAI_ENDPOINT,
        api_version=AZURE_OPENAI_API_VERSION,
    )


def analyze_football_content(transcript: str) -> dict:
    """
    Send transcribed text to Azure OpenAI for football content analysis.

    Args:
        transcript: The transcribed speech text.

    Returns:
        A dict with categorized football data (teams, leagues, sentiment, etc.)
    """
    if not transcript or not transcript.strip():
        return {"error": "ไม่มีข้อความให้วิเคราะห์"}

    client = get_openai_client()

    user_message = FOOTBALL_ANALYSIS_USER_PROMPT.format(transcript=transcript)

    try:
        response = client.chat.completions.create(
            model=AZURE_OPENAI_DEPLOYMENT,
            messages=[
                {"role": "system", "content": FOOTBALL_ANALYSIS_SYSTEM_PROMPT},
                {"role": "user", "content": user_message},
            ],
            max_completion_tokens=4096,
        )

        content = response.choices[0].message.content.strip()

        # Clean up markdown code fences if the model wraps JSON in ```json ... ```
        if content.startswith("```"):
            content = content.split("\n", 1)[1]  # Remove first line (```json)
            content = content.rsplit("```", 1)[0]  # Remove last ```
            content = content.strip()

        result = json.loads(content)
        return result

    except json.JSONDecodeError:
        return {
            "error": "LLM ตอบกลับมาไม่ใช่ JSON ที่ถูกต้อง",
            "raw_response": content,
        }
    except Exception as e:
        return {"error": f"เกิดข้อผิดพลาด: {str(e)}"}


def format_analysis_result(result: dict) -> str:
    """Format the analysis result as a pretty JSON string for display."""
    return json.dumps(result, ensure_ascii=False, indent=2)