app.py

import gradio as gr
import numpy as np
from transformers import pipeline
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import PyPDF2
import torch
import gc

# Load local models for inference
stt_model = pipeline("automatic-speech-recognition", model="openai/whisper-small", torch_dtype=torch.float16)
conversation_model = pipeline("text-generation", model="facebook/blenderbot-400M-distill", torch_dtype=torch.float16)
tts_model = pipeline("text-to-speech", model="facebook/fastspeech2-en-ljspeech", torch_dtype=torch.float16)

# Load a pre-trained model for vector embeddings
embedding_model = SentenceTransformer('all-MiniLM-L6-v2')

# Parse PDF and create resume content
def parse_resume(pdf):
    """Extract text from an uploaded PDF file."""
    reader = PyPDF2.PdfReader(pdf)
    text = "\n".join(page.extract_text() for page in reader.pages if page.extract_text())
    sections = {"Resume Content": text}
    return sections

# Process job description text
def process_job_description(job_desc):
    """Encode the job description for analysis."""
    return embedding_model.encode(job_desc)

# Process resume and generate embeddings
def process_resume(pdf):
    resume_content = parse_resume(pdf)
    resume_embeddings = {
        section: embedding_model.encode(content) for section, content in resume_content.items()
    }
    return resume_embeddings

# Generate question from user response
def generate_question(user_input, resume_embeddings):
    """Find the most relevant section in the resume and generate a question."""
    user_embedding = embedding_model.encode(user_input)
    similarities = {
        section: cosine_similarity([user_embedding], [embedding])[0][0]
        for section, embedding in resume_embeddings.items()
    }
    most_relevant_section = max(similarities, key=similarities.get)
    return f"Based on your experience in {most_relevant_section}, can you elaborate more?"

# Generate TTS output
def generate_audio(text):
    """Convert text to audio using Hugging Face TTS model."""
    audio_data = tts_model(text, return_tensors=True)["waveform"]
    return audio_data

# Gradio interface
class MockInterview:
    def __init__(self):
        self.resume_embeddings = None
        self.job_desc_embedding = None
        self.interview_active = False

    def upload_inputs(self, resume, job_desc):
        self.resume_embeddings = process_resume(resume)
        self.job_desc_embedding = process_job_description(job_desc)
        self.interview_active = True
        question = "Tell me about yourself."
        audio_output = generate_audio(question)
        return "Resume and job description processed. Starting the interview.", audio_output

    def conduct_interview(self, audio_file):
        if not self.interview_active:
            return "Please upload your resume and job description first.", None

        # Transcribe audio
        transcription = stt_model(audio_file)["text"]
        if not transcription.strip():
            return "No audio detected. Please try again.", None

        # Generate next question
        question = generate_question(transcription, self.resume_embeddings)
        audio_output = generate_audio(question)
        return transcription, audio_output

    def end_interview(self):
        self.interview_active = False
        audio_output = generate_audio("Thank you for participating in the interview. Goodbye!")
        return "Interview ended. Thank you for participating.", audio_output

mock_interview = MockInterview()

def upload_inputs(resume, job_desc):
    return mock_interview.upload_inputs(resume, job_desc)

def conduct_interview(audio_file):
    return mock_interview.conduct_interview(audio_file)

def end_interview():
    return mock_interview.end_interview()

interface = gr.Blocks()
with interface:
    gr.Markdown("""# Mock Interview AI
Upload your resume and job description, then engage in a realistic audio-based interview simulation.""")

    with gr.Row():
        resume_input = gr.File(label="Upload Resume (PDF)")
        job_desc_input = gr.Textbox(label="Paste Job Description")
        upload_button = gr.Button("Upload and Start Interview")

    with gr.Row():
        audio_input = gr.Audio(type="filepath", label="Respond with Your Answer")
        transcription_output = gr.Textbox(label="Transcription")
        question_output = gr.Audio(label="Question Audio")
        submit_button = gr.Button("Submit Response")
        end_button = gr.Button("End Interview")

    upload_button.click(upload_inputs, inputs=[resume_input, job_desc_input], outputs=[transcription_output, question_output])
    submit_button.click(conduct_interview, inputs=[audio_input], outputs=[transcription_output, question_output])
    end_button.click(end_interview, outputs=[transcription_output, question_output])

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