app.py

import gradio as gr
import openai
import os
import HongWenData  # Importing the HongWenData module
import base64

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") 
openai.api_key = OPENAI_API_KEY

def image_to_base64(img_path):
    with open(img_path, "rb") as img_file:
        return base64.b64encode(img_file.read()).decode('utf-8')

img_base64 = image_to_base64("HongWenSBC.JPG")
img_html = f'<img src="data:image/jpg;base64,{img_base64}" alt="SBC6" width="300" style="display: block; margin: auto;"/>'

def predict(question_choice, audio):
    # Transcribe the audio using Whisper
    with open(audio, "rb") as audio_file:
        transcript = openai.Audio.transcribe("whisper-1", audio_file)
    message = transcript["text"]  # This is the transcribed message from the audio input
    
    # Generate the system message based on the chosen question
    strategy, explanation = HongWenData.strategy_text["TREES"]

    # Construct the conversation with the system and user's message
    conversation = [
        {
            "role": "system",
            "content": f"""
            You are an expert English Language Teacher in a Singapore Primary school, directly guiding a Primary 6 student in Singapore. 
            The student is answering the question: '{question_choice}'. 
            Point out areas they did well and where they can improve, following the {strategy}. 
            Encourage the use of sophisticated vocabulary and expressions. 
            For the second and third questions, the picture is not relevant, so the student should not refer to it in their response. 
            {explanation} 
            The feedback should be in second person, addressing the student directly.
            """
        },
        {"role": "user", "content": message}
    ]

    response = openai.ChatCompletion.create(
        model='gpt-3.5-turbo',
        messages=conversation,
        temperature=0.6,
        max_tokens=1000,  # Limiting the response to 500 tokens
        stream=True
    )

    partial_message = ""
    for chunk in response:
        if len(chunk['choices'][0]['delta']) != 0:
            partial_message = partial_message + chunk['choices'][0]['delta']['content']
            yield partial_message

# Gradio Interface
iface = gr.Interface(
    fn=predict,
    inputs=[
        gr.Radio(HongWenData.questions, label="Choose a question", default=HongWenData.questions[0]),  # Dropdown for question choice
        gr.inputs.Audio(source="microphone", type="filepath")  # Audio input
    ],
    outputs=gr.inputs.Textbox(),  # Using inputs.Textbox as an output to make it editable
    description=img_html,
    css="custom.css"  # Link to the custom CSS file
)
iface.queue().launch()