app.py

from flask import Flask, request
from flask_cors import CORS
import json

# Loading environment variables
import os
from dotenv import load_dotenv
load_dotenv()
openai_api_key = os.environ.get('openai_api_key')
cohere_api_key = os.environ.get('cohere_api_key')
qdrant_url = os.environ.get('qdrant_url')
qdrant_api_key = os.environ.get('qdrant_api_key')

#Flask config
app = Flask(__name__)
CORS(app)

# Test default route
@app.route('/')
def hello_world():
    return {"Hello":"World"}

## Embedding code
from langchain.embeddings import CohereEmbeddings
from langchain.document_loaders import PyPDFLoader
from langchain.vectorstores import Qdrant

@app.route('/embed', methods=['POST'])
def embed_pdf():
    collection_name = request.json.get("collection_name")
    file_url = request.json.get("file_url")

    loader = PyPDFLoader(file_url)
    docs = loader.load_and_split()
    embeddings = CohereEmbeddings(model="multilingual-22-12", cohere_api_key=cohere_api_key)
    qdrant = Qdrant.from_documents(docs, embeddings, url=qdrant_url, collection_name=collection_name, prefer_grpc=True, api_key=qdrant_api_key)
    
    return {"collection_name":qdrant.collection_name}

# Retrieve information from a collection
from langchain.chains.question_answering import load_qa_chain
from langchain.llms import OpenAI
from qdrant_client import QdrantClient

@app.route('/retrieve', methods=['POST'])
def retrieve_info():
    collection_name = request.json.get("collection_name")
    query = request.json.get("query")

    client = QdrantClient(url=qdrant_url, prefer_grpc=True, api_key=qdrant_api_key)

    embeddings = CohereEmbeddings(model="multilingual-22-12", cohere_api_key=cohere_api_key)
    qdrant = Qdrant(client=client, collection_name=collection_name, embedding_function=embeddings.embed_query)
    search_results = qdrant.similarity_search(query, k=2)
    chain = load_qa_chain(OpenAI(openai_api_key=openai_api_key,temperature=0.2), chain_type="stuff")
    results = chain({"input_documents": search_results, "question": query}, return_only_outputs=True)
    
    return {"results":results["output_text"]}