app.py

import gradio as gr
import whisper
import torch
from transformers import pipeline

# Load Whisper for transcription
asr_model = whisper.load_model("base")

# Load smaller Zephyr model
feedback_model = pipeline(
    "text-generation",
    model="tiiuae/falcon-rw-1b",
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
    device=0 if torch.cuda.is_available() else -1
)

questions = [
    "Why do you want this internship?",
    "What are your strengths and weaknesses?",
    "Tell me about a time you solved a difficult problem.",
    "Where do you see yourself in 5 years?",
    "tell me about yourself",
]

current_question = {"index": 0}

def get_question():
    q = questions[current_question["index"]]
    current_question["index"] = (current_question["index"] + 1) % len(questions)
    return q

def process_audio(audio_file, question):
    """Process the audio file to get feedback."""
    if audio_file is None:
        return "Please record an answer.", ""

    try:
        # Transcribe the audio
        result = asr_model.transcribe(audio_file)
        transcript = result["text"]

        # New prompt with dynamic question
        prompt = (
        f"Interview Question: {question}\n"
        f"Candidate's Answer: {transcript}\n\n"
        "You are an experienced interview evaluator. Review the candidate's response and rate it using the following:\n"
        "- Overall Rating: Weak, Average, Good, or Excellent\n"
        "- Score: Out of 100\n"
        "- Suggestions for improvement (if any)\n\n"
        "Please respond in this format:\n"
        "Overall Rating: <your rating>\n"
        "Score: <your score>\n"
        "Suggestions: <your suggestions>\n"
        )

        # Get feedback from the model
        response = feedback_model(prompt, max_length=200)[0]["generated_text"]

        return response.strip(), transcript

    except Exception as e:
        return f"Error: {str(e)}", ""

with gr.Blocks() as demo:
    gr.Markdown("## AI Interview Coach (Lightweight Version)")

    question_text = gr.Textbox(label="Interview Question", interactive=False)
    get_q_btn = gr.Button("Get Interview Question")

    with gr.Row():
        audio_input = gr.Audio(sources=["microphone"], type="filepath", label="Record Your Answer")
        submit_btn = gr.Button("Submit")

    feedback = gr.Textbox(label="AI Feedback")
    transcript = gr.Textbox(label="Transcript of Your Answer")

    get_q_btn.click(get_question, outputs=question_text)
    submit_btn.click(process_audio, inputs=audio_input, outputs=[feedback, transcript])

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