Update app.py

app.py

@@ -1,6 +1,6 @@
 import gradio as gr
 import time
-from transformers import pipeline
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 from sentence_transformers import SentenceTransformer
 from sklearn.metrics.pairwise import cosine_similarity
 from TTS.api import TTS  # Coqui TTS library
@@ -10,6 +10,9 @@ import PyPDF2
 stt_model = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
 embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
 tts_model = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
+gpt_model_name = "OpenAssistant/oasst-sft-6-llama-30b"
+gpt_tokenizer = AutoTokenizer.from_pretrained(gpt_model_name)
+gpt_model = AutoModelForCausalLM.from_pretrained(gpt_model_name)
 
 # Parse PDF and create resume content
 def parse_resume(pdf):
@@ -26,15 +29,20 @@ def process_inputs(resume, job_desc):
     job_desc_embedding = embedding_model.encode(job_desc)
     return resume_embeddings, job_desc_embedding
 
-# Generate a follow-up question
-def generate_question(response, resume_embeddings):
-    user_embedding = embedding_model.encode(response)
-    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 a follow-up question using GPT
+def generate_question_gpt(response, resume_embeddings, job_description):
+    prompt = f"""
+    You are a hiring manager conducting a professional job interview.
+    Job Description: {job_description}
+    Candidate's Resume Insights: {resume_embeddings}
+    Candidate's Last Response: {response}
+
+    Based on the job description and candidate's resume, generate a professional follow-up question.
+    """
+    inputs = gpt_tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
+    outputs = gpt_model.generate(**inputs, max_length=150, num_beams=3, early_stopping=True)
+    question = gpt_tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return question.strip()
 
 # Generate TTS audio for a question
 def generate_audio(question):
@@ -66,8 +74,8 @@ class MockInterview:
         if not transcription.strip():
             return "No response detected. Please try again.", None
 
-        # Generate the next question
-        self.current_question = generate_question(transcription, self.resume_embeddings)
+        # Generate the next question using GPT
+        self.current_question = generate_question_gpt(transcription, self.resume_embeddings, self.job_desc_embedding)
         return transcription, generate_audio(self.current_question)
 
     def end_interview(self):
@@ -98,11 +106,11 @@ with interface:
     audio_input = gr.Audio(type="filepath", label="Your Response")
     question_audio_output = gr.Audio(label="Question Audio")
     transcription_output = gr.Textbox(label="Transcription")
-    interaction_button = gr.Button("Next Interaction")
-    end_button = gr.Button("End Interview")
 
-    resume_uploaded = resume_input.change(start_interview, inputs=[resume_input, job_desc_input], outputs=[transcription_output, question_audio_output])
-    interaction_button.click(next_interaction, inputs=[audio_input], outputs=[transcription_output, question_audio_output])
+    resume_input.change(start_interview, inputs=[resume_input, job_desc_input], outputs=[transcription_output, question_audio_output])
+    audio_input.change(next_interaction, inputs=[audio_input], outputs=[transcription_output, question_audio_output])
+
+    end_button = gr.Button("End Interview")
     end_button.click(end_interview, outputs=[transcription_output, question_audio_output])
 
 if __name__ == "__main__":