app.py

import os
import time
import requests
from flask import Flask, request, jsonify
from flask_cors import CORS

import openai
import langchain
import random
from langchain_openai import ChatOpenAI
from langchain_community.cache import InMemoryCache
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

# Set environment variables and OpenAI configurations
api_keys = os.environ
openai.api_key = os.environ['OPENAI_API_KEY']
print(os.environ['OPENAI_API_KEY'])
langchain.llm_cache = InMemoryCache()

app = Flask(__name__)
CORS(app)

# This sends final LLM output to dynamically get tagged with UneeQ emotions and expressions
def process_emotions(query):
    try:

        # URL of the FT LLM API endpoint (ECS)
        url = "https://api-ft-inline-tags.caipinnovation.com/query"

        # Payload to be sent in the POST request
        payload = {"prompt": query}

        # Making the POST request
        response = requests.post(url, json=payload)

        # Checking if the request was successful
        response.raise_for_status()

        # Returning the 'answer' from the response
        ft_answer = response.json().get("answer")
        return ft_answer

    except Exception as e:
        raise 

#This handles general Q&A to the LLM
def process_query(query, chat_history, systemMessage, emotions):
    try:

        print(f"calling fine_tuned_model") 

        #Get name of model from ENV
        ft_model_name = os.environ.get("OPENAI_MODEL_NAME")
        
        #Model name from env will be used here:
        fine_tuned_model = ChatOpenAI(
            temperature=0, model_name=ft_model_name
        )

        prompt_template = """System: {systemMessage}.
        User: The user is inquiring about cataracts or cataract surgery.  Answer their question: {query}"""
        PROMPT = PromptTemplate(template=prompt_template,
                                input_variables=["systemMessage", "query"])

        chain = LLMChain(llm=fine_tuned_model, prompt=PROMPT, verbose=False)

        input_prompt = [{"systemMessage": systemMessage, "query": query}]

        generatedResponse = chain.apply(input_prompt)

        #Replace/filter out any prepended strings from LLM response
        #Sometimes we have issues that the LLM writes these following strings before answer. Use if needed.
        llm_response = generatedResponse[0]["text"].replace("Answer:", "").replace("System:", "").lstrip()
        print(llm_response)

        #NOW SEND RESPONSE TO GET TAGGED w/ Emotions and Expressions
        if emotions:
            try:
                llm_response_ft = process_emotions(llm_response)
            except Exception as e:
                # Log the error
                print(f"Error processing emotions for query: {llm_response}. Error: {str(e)}")
                # Return the error response
                return {"error": "Error processing emotions", "query": llm_response}

            return {
                "answer": llm_response_ft,
                "source_documents": ""
            }
        else:
            return {
                "answer": llm_response,
                "source_documents": ""
            }
    except Exception as e:
        print(f"Error processing query: {query}. Error: {str(e)}")
        return {"error": "Error processing query"}
    
#This handles the chart functionality in HIMSS    
def process_chart(query, s1 ,s2):
    try:
        print("calling fine_tuned_model")    
        #Get name of model from ENV
        ft_model_name = os.environ.get("OPENAI_MODEL_NAME")
        
        #Model name from env will be used here:
        fine_tuned_model = ChatOpenAI(
            temperature=0, model_name=ft_model_name
        )

        prompt_template = """System: {systemMessage}.
        User: {query}"""
        PROMPT = PromptTemplate(template=prompt_template,
                                input_variables=["systemMessage", "query"])

        chain = LLMChain(llm=fine_tuned_model, prompt=PROMPT, verbose=False)

        # Get systemMessage from env file:
        systemMessage = os.environ.get("SYSTEM_MESSAGE")

        input_prompt = [{"systemMessage": systemMessage, "query": query}]

        generatedResponse = chain.apply(input_prompt)
        print("after ", generatedResponse[0]["text"])

        #Replace/filter out any prepended strings from LLM response
        #Sometimes we have issues that the LLM writes these following strings before answer. Use if needed.
        generatedResponse_filtered = generatedResponse[0]["text"].replace("Answer:", "").replace("System:", "").lstrip()

        stripped_answer = f"I see you have {s1} and {s2}. {generatedResponse_filtered}"

        return {
            "answer": stripped_answer,
            "source_documents": ""
        }
    except Exception as e:
        print(f"Error processing query: {query}. Error: {str(e)}")
        return {"error": "Error processing query"}

#POST request to this service
@app.route('/query', methods=['POST'])
def handle_query():

    data = request.json
    query=data['prompt']
    chatHistory=data['chatHistory']
    systemMessage='You are a helpful medical assistant.'
    answer = ''
    emotions = ''

    result = process_query(query, chatHistory, systemMessage, emotions)
    answer = result['answer']

    serialized_result = {
        "answer": answer,
        "matchedContext": "",
        "conversationPayload": ""
    }

    print(serialized_result)
    
    return jsonify(serialized_result), 200

# Helper Functions
def pick_random_issues(issues):     
    # Randomly select two strings from the list
    random_strings = random.sample(issues, 2)
    return random_strings

def generate_description():
    issues = ["Anterior Uveitis", "Corneal Guttata", "Diabetes", "Diabetes Mellitus", "Glaucoma", "Retinal Detachment", "Corticosteroids", "Phenothiazine", "Chlorpromazine", "Ultraviolet Radiation Exposure", "Smoking", "High Alcohol Consumption", "Poor Nutrition"]
    random_strings = pick_random_issues(issues)
    random_string_1, random_string_2 = random_strings
    description = f"Describe any issues I may encounter due to {random_string_1} and {random_string_2} relative to my upcoming cataract surgery?"
    return description, random_string_1, random_string_2

#GET request to chart feature
@app.route('/chart', methods=['GET'])
def handle_chart():
    description, random_string_1, random_string_2 = generate_description()
    query = description
    if not query:
        return jsonify({"error": "No query provided"}), 400
    result = process_chart(query, random_string_1, random_string_2)
    if "error" in result:
        return jsonify(result), 500

    serialized_result = {
        "query": query,
        "answer": result["answer"],
        "source_documents": ""
    }
    
    return jsonify(serialized_result), 200


@app.route('/')
def hello():
    version = os.environ.get("CODE_VERSION")
    return jsonify({"status": "Healthy", "version": version}), 200


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