api.py

from flask import Flask, request, jsonify
from langchain_community.vectorstores import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_ollama import OllamaLLM
from langchain.chains.question_answering import load_qa_chain
from langchain import PromptTemplate

app = Flask(__name__)

# Initialize the language models
llm = OllamaLLM(model="llama3.2")

# Initialize HuggingFaceEmbeddings and Chroma
model_name = "intfloat/multilingual-e5-large"
load_from_dir = "Hadith_Chroma_db"

embedding_llm = HuggingFaceEmbeddings(model_name=model_name)

loaded_vector_db = Chroma(
    persist_directory=load_from_dir,
    embedding_function=embedding_llm
)

def get_similar_docs(query):
    """Retrieve similar documents based on the query."""
    similar_docs = loaded_vector_db.similarity_search(query, k=2)
    return similar_docs

def ask_llms(query_text):
    """Ask the LLM to provide an answer based on similar documents."""
    similar_docs = get_similar_docs(query_text)

    qna_template = '\n'.join([
        "Answer the following question using the context provided.",
        "If the answer is not included in the context, say 'No answer available'.",
        "### Context:",
        "{context}",
        "### Question:",
        "{question}",
        "### Answer:"
    ])

    qna_prompt = PromptTemplate(
        template=qna_template,
        input_variables=['context', 'question'],
        verbose=True
    )

    stuff_chain = load_qa_chain(llm, chain_type="stuff", prompt=qna_prompt)

    final_answer = stuff_chain.invoke({
        "input_documents": similar_docs,
        "question": query_text
    })

    return final_answer['output_text']

@app.route('/ai', methods=['POST'])
def aiPost():
    """Handle POST requests to the /ai endpoint."""
    try:
        json_content = request.json
        if not json_content or 'query' not in json_content:
            return jsonify({"error": "Invalid input, 'query' field is required"}), 400
        
        query = json_content.get('query')
        
        # Get the response from the LLM based on the query
        response = ask_llms(query)
        
        return jsonify({"response": response})
    
    except Exception as e:
        return jsonify({"error": str(e)}), 500

def start_app():
    """Start the Flask app."""
    app.run(host="0.0.0.0", port=8080, debug=True)

if __name__ == '__main__':
    start_app()