LLM.py

from openai import OpenAI
from config import (
    gemini_api_key_1, gemini_api_key_2, gemini_api_key_3, gemini_api_key_4,
    query_gen_api_key_1, query_gen_api_key_2, query_gen_api_key_3, query_gen_api_key_4
)



from typing import List
client_1 = OpenAI(
    api_key=gemini_api_key_1,
    base_url="https://generativelanguage.googleapis.com/v1beta/"
)
client_2 = OpenAI(
    api_key=gemini_api_key_2,
    base_url="https://generativelanguage.googleapis.com/v1beta/"
)
client_3 = OpenAI(
    api_key=gemini_api_key_3,
    base_url="https://generativelanguage.googleapis.com/v1beta/"
)
client_4 = OpenAI(
    api_key=gemini_api_key_4,
    base_url="https://generativelanguage.googleapis.com/v1beta/"
)

# Query generation clients
query_client_1 = OpenAI(
    api_key=query_gen_api_key_1,
    base_url="https://api.groq.com/openai/v1"
)
query_client_2 = OpenAI(
    api_key=query_gen_api_key_2,
    base_url="https://api.groq.com/openai/v1"
)
query_client_3 = OpenAI(
    api_key=query_gen_api_key_3,
    base_url="https://api.groq.com/openai/v1"
)
query_client_4 = OpenAI(
    api_key=query_gen_api_key_4,
    base_url="https://api.groq.com/openai/v1"
)

def get_chat_completion(query: str, client) -> dict:
        response = client.chat.completions.create(
            model="gemini-2.5-flash",  # Or any other model you want to use
            n=1,
            messages=[
                {"role": "system", "content": "You are an assistant that answers questions only using the given context. Do not guess or add anything outside the context.  "},
                {"role": "user", "content": query}
            ]
        )
        return response.choices[0].message.content

def construct_prompts(questions: List[str], contexts: List[str]) -> List[str]:
    return [
        f"CONTEXT:\n{context}\n\nQUESTION:\n{question}"
        for context, question in zip(contexts, questions)
    ]
import concurrent.futures

gemini_clients = [client_1, client_2, client_3, client_4]

def generate_answers(prompts: List[str], clients: List = gemini_clients) -> List[str]:
    answers = [None] * len(prompts)  # Will fill answers by index

    def gemini_task(idx_client_prompt):
        idx, client, prompt = idx_client_prompt
        ans = get_chat_completion(prompt, client)
        return idx, ans

    tasks = [
        (i, gemini_clients[i % len(gemini_clients)], prompts[i])
        for i in range(len(prompts))
    ]

    with concurrent.futures.ThreadPoolExecutor(max_workers=len(gemini_clients)) as executor:
        # Ensure order by writing result at respective index
        for idx, ans in executor.map(gemini_task, tasks):
            answers[idx] = ans

    return answers

# Query generation functions
query_gen_clients = [query_client_1, query_client_2, query_client_3, query_client_4]

def get_query_generation(user_input: str, client) -> str:
    """Generate search queries based on user input using query generation clients."""
    response = client.chat.completions.create(
        model="meta-llama/llama-4-scout-17b-16e-instruct",
        n=1,
        messages=[
            {"role": "system", "content": "You are an expert at generating search queries. Given a user input, generate the most relevant search query that would help find the best information to answer the user's question. Return only the search query, nothing else."},
            {"role": "user", "content": f"Generate a search query for: {user_input}"}
        ]
    )
    return response.choices[0].message.content

def generate_search_queries(user_inputs: List[str], clients: List = query_gen_clients) -> List[str]:
    """Generate search queries for multiple user inputs using parallel processing."""
    queries = [None] * len(user_inputs)

    def query_gen_task(idx_client_input):
        idx, client, user_input = idx_client_input
        query = get_query_generation(user_input, client)
        return idx, query

    tasks = [
        (i, query_gen_clients[i % len(query_gen_clients)], user_inputs[i])
        for i in range(len(user_inputs))
    ]

    with concurrent.futures.ThreadPoolExecutor(max_workers=len(query_gen_clients)) as executor:
        for idx, query in executor.map(query_gen_task, tasks):
            queries[idx] = query

    return queries

def process_user_queries_with_context(user_inputs: List[str]) -> List[str]:
    from qdrant_setup import get_context_for_questions

    search_queries = generate_search_queries(user_inputs)
    contexts = get_context_for_questions(search_queries)
    prompts = construct_prompts(user_inputs, contexts)
    answers = generate_answers(prompts)
    return answers

if __name__=="__main__":
   # Example user questions
   user_questions = [
         "What is the grace period for premium payment under the National Parivar Mediclaim Plus Policy?",
         "What is the waiting period for pre-existing diseases (PED) to be covered?",
         "Does this policy cover maternity expenses, and what are the conditions?",
         "What is the waiting period for cataract surgery?"
   ]
   process_user_queries_with_context(user_questions)
   
#    print("=== Testing complete pipeline ===")
#    # Test the complete pipeline with query generation
#    answers = process_user_queries_with_context(user_questions)
#    for i, (question, answer) in enumerate(zip(user_questions, answers)):
#        print(f"\nQ{i+1}: {question}")
#        print(f"A{i+1}: {answer}")
   
#    print("\n=== Testing original approach ===")
#    # Original approach for comparison
#    from qdrant_setup import get_context_for_questions
#    context = get_context_for_questions(user_questions)
#    prompts = construct_prompts(user_questions, context)
#    original_answers = generate_answers(prompts)
   
#    print("\nComparison:")
#    for i, (orig, new) in enumerate(zip(original_answers, answers)):
#        print(f"\nQuestion {i+1}:")
#        print(f"Original: {orig[:100]}...")
#        print(f"New: {new[:100]}...")