llm_handler.py

# rentbot/llm_handler.py
import os
from openai import AsyncOpenAI

client = AsyncOpenAI(api_key=os.getenv("OPENAI_API_KEY"))

# Definition of the tool the LLM can use
tools = [
    {
        "type": "function",
        "function": {
            "name": "create_event",
            "description": "Create a calendar event to book an apartment viewing.",
            "parameters": {
                "type": "object",
                "properties": {
                    "start_time": {
                        "type": "string",
                        "description": "The start time of the event in ISO 8601 format, e.g., 2025-07-18T14:00:00",
                    },
                    "duration_minutes": {
                        "type": "integer",
                        "description": "The duration of the event in minutes.",
                        "default": 30
                    },
                    "summary": {
                        "type": "string",
                        "description": "A short summary or name for the event, e.g., 'Unit 5B viewing'",
                    },
                },
                "required": ["start_time", "summary"],
            },
        },
    }
]

async def get_llm_response(messages: list):
    """
    Yields response chunks from OpenAI's chat completion streaming API.
    Returns the full assistant message and any tool calls.
    """
    try:
        stream = await client.chat.completions.create(
            model="gpt-4o-mini",
            messages=messages,
            stream=True,
            tools=tools,
            tool_choice="auto",
        )
        
        full_response = ""
        tool_calls = []
        
        async for chunk in stream:
            delta = chunk.choices[0].delta
            if delta.content:
                text_chunk = delta.content
                full_response += text_chunk
                yield text_chunk # Yield text chunks for real-time TTS
            
            if delta.tool_calls:
                # Accumulate tool call chunks
                if not tool_calls:
                    tool_calls.extend(delta.tool_calls)
                else:
                    for i, tool_call_chunk in enumerate(delta.tool_calls):
                        if tool_call_chunk.function.arguments:
                            tool_calls[i].function.arguments += tool_call_chunk.function.arguments

        # Construct the final assistant message object
        assistant_message = {"role": "assistant", "content": full_response}
        if tool_calls:
            assistant_message["tool_calls"] = [
                {
                    "id": tc.id,
                    "type": "function",
                    "function": {"name": tc.function.name, "arguments": tc.function.arguments}
                } for tc in tool_calls
            ]
            
        return assistant_message, tool_calls

    except Exception as e:
        print(f"Error in get_llm_response: {e}")
        return {"role": "assistant", "content": "I'm having a little trouble right now. Please try again in a moment."}, []