app.py

from flask import Flask, request, jsonify, render_template
from openai import OpenAI
import requests
import sentiment_text_helpers
from subprocess import call
import os
from dotenv import load_dotenv

load_dotenv()

app = Flask(__name__)
openai_client = OpenAI()

def query(payload):
    hf_api_url = "https://api-inference.huggingface.co/models/SamLowe/roberta-base-go_emotions"
    headers = {"Authorization": "Bearer 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"}

    response = requests.post(hf_api_url, headers=headers, json=payload)
    return response.json()


def generate_questions(input_prompt):
    with open('template.txt', 'r') as file:
        template = file.read()

    # when user enters a new sentence, they won't type "Sentence: " first, so we do that
    input_prompt = 'You are interviewing a candidate for a software engineering job in a behavioral interview. Generate responses and interview questions based on the given conversation context. Sentence: ' + input_prompt
    # pass the input prompt AND the template into the completions create function
    prompt = template + input_prompt
    # more tokens means longer responses, higher temperature means more creative responses
    completion = openai_client.completions.create(model='gpt-3.5-turbo-instruct', prompt=str(prompt), max_tokens=128, temperature=0.7)
    print(completion)
    # extract the text section of completion object
    message = completion.choices[0].text
    
    output_list = message.split('\n')
    out_index = []
    questions = None
    for idx, sentence in enumerate(output_list):
        if 'Question' in sentence:
            out_index.append(idx)
    if out_index:
        questions = output_list[min(out_index):max(out_index) + 1]

    if not questions:
        final_output = "Can you tell us about a time when you had to balance user needs with technical constraints while developing an application?"
    else:
        final_output = ""
        for question in questions[:2]:
            final_output += f"{question[10:]}\n"
    return final_output


# homepage
@app.route('/')
def index():
    return render_template('index.html')


@app.route('/whisper', methods=['POST']) 
def transcribe_audio():
    """Takes a file path for an audio file and transcribes it into a string"""
    
    # open conversation history and record the last couple things said
    prior = ""
    with open("conversation_history.txt", 'r') as file:
        prior = file.read()

    if 'audio_data' not in request.files:
        return jsonify({'message': 'No file part'}), 400

    audio_file = request.files['audio_data']

    # initially save to webm (that's the type it's sent in too)
    audio_file.save(dst='audio.webm')

    # convert webm to wav using call function
    call(['ffmpeg', '-i', 'audio.webm', '-ar', '16000', '-ac', '1', 'audio.wav'])

    with open('audio.wav', 'rb') as audio_file:
        transcript = openai_client.audio.transcriptions.create(
            model="whisper-1",
            file=audio_file,
        )
    
    file = open("conversation_history.txt", "a")
    file.write(f"{transcript.text}\n")
    this_answer_transcript = transcript.text

    prior_and_transcript = prior + this_answer_transcript

    os.remove('audio.webm')
    os.remove('audio.wav')

    # run question gen on transcript
    questions = f"Question: {generate_questions(prior_and_transcript)}"

    # run tone analysis
    tone_single = sentiment_text_helpers.give_sentiment_question(this_answer_transcript)
    tone_analyzis = f"Feedback: {tone_single}"

    # open convo history file to write to it
    file.write(f"\n{questions}\n")
    file.close()

    return jsonify({'message': 'Transcript received', 'transcript': this_answer_transcript, 'questions': questions, 'tone analysis': tone_analyzis})


@app.route('/finaltone', methods=['POST']) 
def return_tone_results():
    with open("conversation_history.txt", 'r') as file:
        transcript = file.read()
    
    return jsonify({'message': "End of interview data processing.", 'sentiment-results': sentiment_text_helpers.give_sentiment_full(transcript)})



def main():
    with open("conversation_history.txt", 'w'):
        pass
    app.run(port=6060, debug=True)


if __name__ == '__main__':
    main()