app.py

import os

CHROMA_PATH = '/code/chroma_db'
if not os.path.exists(CHROMA_PATH):
    os.makedirs(CHROMA_PATH)
from langchain.vectorstores.chroma import Chroma
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter

# Load and process the PDF
def save_pdf_and_update_database(pdf_filepath):
    try:
        # Load the PDF
        document_loader = PyPDFLoader(pdf_filepath)
        documents = document_loader.load()
        
        # Split the documents into manageable chunks
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size=800,
            chunk_overlap=80,
            length_function=len,
            is_separator_regex=False,
        )
        chunks = text_splitter.split_documents(documents)
        
        # Initialize Chroma with an embedding function
        embedding_function = HuggingFaceEmbeddings()
        db = Chroma(persist_directory=CHROMA_PATH, embedding_function=embedding_function)
        
        # Add chunks to ChromaDB
        db.add_documents(chunks)
        db.persist()
        print("PDF processed and data updated in Chroma.")
    except Exception as e:
        print(f"Error processing PDF: {e}")

AI71_API_KEY = os.environ.get('AI71_API_KEY')

def generate_response(query, chat_history):
    response = ''
    for chunk in AI71(AI71_API_KEY).chat.completions.create(
            model="tiiuae/falcon-180b-chat",
            messages=[
                {"role": "system", "content": "You are the best agricultural assistant. Remember to give a response in not more than 2 sentences."},
                {"role": "user", "content": f'''Answer the query based on history {chat_history}: {query}'''},
            ],
            stream=True,
    ):
        if chunk.choices[0].delta.content:
            response += chunk.choices[0].delta.content
    return response.replace("###", '').replace('\nUser:', '')

def query_rag(query_text: str, chat_history):
    db = Chroma(persist_directory=CHROMA_PATH, embedding_function=HuggingFaceEmbeddings())
    
    # Perform a similarity search in ChromaDB
    results = db.similarity_search_with_score(query_text, k=5)
    
    if not results:
        return "Sorry, I couldn't find any relevant information."
    
    context_text = "\n\n---\n\n".join([doc.page_content for doc, _score in results])
    
    # Generate the response using the Falcon model
    prompt = f"Context:\n{context_text}\n\nQuestion:\n{query_text}"
    response = generate_response(prompt, chat_history)
    
    return response


@app.route('/whatsapp', methods=['POST'])
def whatsapp_webhook():
    incoming_msg = request.values.get('Body', '').lower()
    sender = request.values.get('From')
    num_media = int(request.values.get('NumMedia', 0))

    chat_history = conversation_memory.get_memory()

    if num_media > 0:
        media_url = request.values.get('MediaUrl0')
        content_type = request.values.get('MediaContentType0')

        if content_type == 'application/pdf':
            # Handle PDF processing
            filepath = download_file(media_url, ".pdf")
            save_pdf_and_update_database(filepath)
            response_text = "PDF has been processed. You can now ask questions related to its content."
        else:
            response_text = "Unsupported file type. Please upload a PDF document."
    else:
        # Handle queries
        response_text = query_rag(incoming_msg, chat_history)

    conversation_memory.add_to_memory({"user": incoming_msg, "assistant": response_text})
    send_message(sender, response_text)
    return '', 204


if __name__ == "__main__":
    send_initial_message('919080522395')
    send_initial_message('916382792828')
    app.run(host='0.0.0.0', port=7860)