agents.py

import os
import json
import time
import traceback
from dotenv import load_dotenv

from langfuse import Langfuse
from langchain_groq import ChatGroq
from langfuse.callback import CallbackHandler
from langchain_core.prompts import PromptTemplate

load_dotenv()

langfuse_news_analysis_handler = CallbackHandler(
    secret_key=os.getenv("LANGFUSE_SECRET_KEY"),
    public_key=os.getenv("LANGFUSE_PUBLIC_KEY"),
    host="https://cloud.langfuse.com", # 🇪🇺 EU region
    session_id="news_analysis",
)
langfuse_post_generation_handler = CallbackHandler(
    secret_key=os.getenv("LANGFUSE_SECRET_KEY"),
    public_key=os.getenv("LANGFUSE_PUBLIC_KEY"),
    host="https://cloud.langfuse.com", # 🇪🇺 EU region
    session_id="post_generation",
)
langfuse = Langfuse()

analysis_llm = ChatGroq(
    model="llama-3.1-8b-instant",
    temperature=0.8,
    timeout=None,
    max_retries=2,
    api_key=os.getenv("GROQ_ANALYSIS_API_KEY"),
)
post_content_llm = ChatGroq(
    model="qwen-qwq-32b",
    temperature=0.8,
    timeout=None,
    max_retries=2,
    api_key=os.getenv("THREADS_POST_GENERATION_API_KEY"),
)

    
def basic_analysis(news):
    prompt = langfuse.get_prompt("news_selector")

    for _ in range(5):
        try:
            response = analysis_llm.invoke(
                prompt.compile(news_object = news), 
                config={"callbacks": [langfuse_news_analysis_handler]}
            )
            print("################ BASIC ANALYSIS AGENT RESPONSE ################")
            print(response.content)
            print("################ BASIC ANALYSIS END AGENT RESPONSE ################")

            if "</think>" in response.content:
                response.content = response.content.split("</think>")[1]

            start_index = response.content.find("{")
            end_index = response.content.rfind("}")

            print("start index:", start_index)
            print("end index:", end_index)

            abstracted_string = ""
            if start_index != -1 and end_index != -1 and start_index < end_index:
                abstracted_string = response.content[start_index : end_index + 1]

                try:
                    results = json.loads(abstracted_string)
                    print(results)
                    return results
                except Exception as e:
                    print(e)
                    traceback.print_exc()
        except Exception as e:
            print(e)
            traceback.print_exc()
            time.sleep(30)

    return {"error": "LLM response is not in correct format."}

def get_text_post_content(details, reference):
    try:
        prompt = PromptTemplate.from_file(
            template_file="prompts/post_generator_without_source.yml",
            input_variables=["NEWS_CONTENT", "CHAR_LENGTH"],
        )
        prompt = langfuse.get_prompt("post_generator")


        user_query = prompt.compile(NEWS_CONTENT = details, CHAR_LENGTH = 490- len(reference))
        response = post_content_llm.invoke(user_query, config={"callbacks": [langfuse_post_generation_handler]})

        print("POST CONTENT RESPONSE:", response)
        
        content = response.content.replace('"', '')

        if "</think>" in content:
            content = content.split("</think>")[1]

        start_indx = content.find("#")
        content = f"""{content[:start_indx]}
{reference}

{content[start_indx:]}"""
        
        return content, True
    except Exception as e:
        print(e)
        traceback.print_exc()
        return "", False