app.py

import time
import gradio as gr
import os
import asyncio
from pymongo import MongoClient
from langchain_community.vectorstores import MongoDBAtlasVectorSearch
from langchain_openai import OpenAIEmbeddings
from langchain_community.llms import OpenAI
# from langchain_community.prompts import PromptTemplate
# from langchain.chains import LLMChain
import json


## Connect to MongoDB Atlas local cluster
MONGODB_ATLAS_CLUSTER_URI = os.getenv('MONGODB_ATLAS_CLUSTER_URI')
client = MongoClient(MONGODB_ATLAS_CLUSTER_URI)
db_name = 'sample_mflix'
collection_name = 'embedded_movies'
collection = client[db_name][collection_name]

## Create a vector search index
print ('Creating vector search index')
# collection.create_search_index(model={"definition": {"mappings":{
#   "dynamic":True,
#   "fields": {
#     "plot_embedding": {
#       "type": "knnVector",
#       "dimensions": 1536,
#       "similarity": "euclidean"
#     }
#   }
# }}, "name":'default'})

# sleep for minute
# print ('Waiting for vector index on field "embedding" to be created')
# time.sleep(60)

vector_store = MongoDBAtlasVectorSearch(embedding=OpenAIEmbeddings(), collection=collection, index_name='vector_index', text_key='plot', embedding_key='plot_embedding')

def get_movies(message, history):
    # Use AsyncIO to run the similarity search in the background
   # movies =  vector_store.similarity_search(message, 3)
    print ('Searching for: ' + message)
    movies =  vector_store.similarity_search(message, 3)
    retrun_text = ''
    for movie in movies:
        retrun_text = retrun_text + 'Title : ' +  movie.metadata['title'] + '\n------------\n' + 'Plot: ' + movie.page_content + '\n\n'

    for i in range(len(retrun_text)):
        time.sleep(0.05)
        yield "Found: " + "\n\n" + retrun_text[: i+1]
    

demo = gr.ChatInterface(get_movies, examples=["What movies are scary?", "Find me a comedy", "Movies for kids"], title="Movies Atlas Vector Search", submit_btn="Search").queue()

if __name__ == "__main__":
    demo.launch()