# --- IMPORTS --- import gradio as gr from huggingface_hub import InferenceClient import re import random import whisper from pydub import AudioSegment import torch import torchvision.transforms as transforms import torchvision.models as models from PIL import Image # --- LOAD MODELS --- # HuggingFace Zephyr Model for Chat client = InferenceClient("HuggingFaceH4/zephyr-7b-beta") # Whisper Model for Audio-to-Text whisper_model = whisper.load_model("base") # ResNet18 Model for Posture Classification model = models.resnet18(weights=models.ResNet18_Weights.IMAGENET1K_V1) model.fc = torch.nn.Linear(model.fc.in_features, 2) model.eval() transform = transforms.Compose([ transforms.Resize((224, 224)), transforms.ToTensor() ]) # --- LOAD QUESTIONS --- def load_questions(file_path): with open(file_path, 'r') as f: data = f.read() question_blocks = re.split(r'Question:\s*', data)[1:] questions = [] for block in question_blocks: parts = block.split('Possible Answers:') question_text = parts[0].strip() answers_text = parts[1].strip() possible_answers = [ans.strip() for ans in re.split(r'\d+\.\s+', answers_text) if ans.strip()] questions.append({'question': question_text, 'answers': possible_answers}) return questions all_questions = load_questions('knowledge.txt') questions_by_type = { 'Technical': [q for q in all_questions if any(keyword in q['question'].lower() for keyword in [ 'function', 'linked list', 'url', 'rest', 'graphql', 'garbage', 'cap theorem', 'sql', 'hash table', 'stack', 'queue', 'recursion', 'reverse', 'bfs', 'dfs', 'time complexity', 'binary search tree', 'web application', 'chat system', 'load balancing', 'caching', 'normalization', 'acid', 'indexing', 'sql injection', 'https', 'xss', 'hash', 'vulnerabilities'])], 'Competency-Based Interview': [q for q in all_questions if any(keyword in q['question'].lower() for keyword in [ "debugging", "learning", "deadlines", "teamwork", "leadership", "mistake", "conflict", "decision"])], 'Case': [q for q in all_questions if any(keyword in q['question'].lower() for keyword in [ "testing", "financial", "automation", "analysis", "regression", "business", "stakeholder"])] } # --- AUDIO TRANSCRIPTION --- def transcribe_audio(file_path): try: audio = AudioSegment.from_file(file_path) converted_path = "converted.wav" audio.export(converted_path, format="wav") result = whisper_model.transcribe(converted_path, fp16=False) return result["text"] except Exception as e: return f"❌ ERROR: {str(e)}" # --- POSTURE CLASSIFICATION --- def classify_image(image): if image is None: return "No image provided! Please upload or capture an image." image = transform(image).unsqueeze(0) output = model(image) _, predicted = torch.max(output, 1) return ( "✅ Good Posture! Sit exactly like that for your Interview!" if predicted.item() == 0 else "⚠️ Bad Posture — try sitting straighter or more centered for your real interview." ) # --- INTERVIEW LOGIC --- def set_type(choice, user_profile): user_profile["interview_type"] = choice return "Great! What’s your background and what field/role are you aiming for?", user_profile def save_background(info, user_profile): user_profile["field"] = info return "Awesome! Type 'start' below to begin your interview.", user_profile def respond(message, chat_history, user_profile): message_lower = message.strip().lower() if not user_profile.get("interview_type") or not user_profile.get("field"): bot_msg = "Please finish steps 1 and 2 before starting the interview." chat_history.append((message, bot_msg)) return chat_history if message_lower == 'start': interview_type = user_profile['interview_type'] selected_questions = questions_by_type.get(interview_type, []) random.shuffle(selected_questions) selected_questions = selected_questions[:10] user_profile['questions'] = selected_questions user_profile['current_q'] = 0 user_profile['user_answers'] = [] user_profile['interview_in_progress'] = True intro = f"Welcome to your {interview_type} interview for a {user_profile['field']} position. I will ask you up to 10 questions. Type 'stop' anytime to end." first_q = f"First question: {selected_questions[0]['question']}" chat_history.append((message, intro)) chat_history.append(("", first_q)) return chat_history if message_lower == 'stop' and user_profile.get("interview_in_progress"): user_profile['interview_in_progress'] = False bot_msg = "Interview stopped. Type 'feedback' if you'd like me to analyze your answers. Thanks for interviewing with Intervu!" chat_history.append((message, bot_msg)) return chat_history if user_profile.get("interview_in_progress"): q_index = user_profile['current_q'] user_profile['user_answers'].append(message) q_index += 1 user_profile['current_q'] = q_index if q_index < len(user_profile['questions']): bot_msg = f"Next question: {user_profile['questions'][q_index]['question']}" else: user_profile['interview_in_progress'] = False bot_msg = "Interview complete! Type 'feedback' if you'd like me to analyze your answers. Thanks for interviewing with Intervu!" chat_history.append((message, bot_msg)) return chat_history if message_lower == 'feedback': feedback = generate_feedback(user_profile) chat_history.append((message, feedback)) return chat_history messages = [{"role": "system", "content": f"You are a professional interviewer conducting a {user_profile['interview_type']} interview for a candidate in {user_profile['field']}."}] for q, a in chat_history: messages.append({"role": "user", "content": q}) messages.append({"role": "assistant", "content": a}) messages.append({"role": "user", "content": message}) response = client.chat_completion(messages, max_tokens=150, stream=False) bot_msg = response.choices[0].message.content chat_history.append((message, bot_msg)) return chat_history def generate_feedback(user_profile): feedback = [] questions = user_profile.get('questions', []) answers = user_profile.get('user_answers', []) for i, user_ans in enumerate(answers): correct_answers = questions[i]['answers'] match = any(ans.lower() in user_ans.lower() for ans in correct_answers) if match: fb = f"Question {i+1}: ✅ Good job!" else: fb = f"Question {i+1}: ❌ Missed key points: {correct_answers[0]}" feedback.append(fb) return "\n".join(feedback) # AUDIO HANDLING def handle_audio(audio_file, chat_history, user_profile): transcribed = transcribe_audio(audio_file) if transcribed.startswith("❌"): chat_history.append(("Audio input", transcribed)) return chat_history return respond(transcribed, chat_history, user_profile) # --- GRADIO INTERFACE --- with gr.Blocks() as demo: user_profile = gr.State({"interview_type": "", "field": "", "interview_in_progress": False}) chat_history = gr.State([]) gr.Markdown("# 🎤 Welcome to Intervu") gr.Image(value="images.JPEG", show_label=False, width=200) gr.Markdown("### Step 1: Choose Interview Type") with gr.Row(): with gr.Column(): btn1 = gr.Button("Technical") btn2 = gr.Button("Competency-Based Interview") btn3 = gr.Button("Case") type_output = gr.Textbox(label="Bot response", interactive=False) btn1.click(set_type, inputs=[gr.Textbox(value="Technical", visible=False), user_profile], outputs=[type_output, user_profile]) btn2.click(set_type, inputs=[gr.Textbox(value="Competency-Based Interview", visible=False), user_profile], outputs=[type_output, user_profile]) btn3.click(set_type, inputs=[gr.Textbox(value="Case", visible=False), user_profile], outputs=[type_output, user_profile]) gr.Markdown("### Step 2: Enter Your Background") background = gr.Textbox(label="Your background and field/goal") background_btn = gr.Button("Submit") background_output = gr.Textbox(label="Bot response", interactive=False) background_btn.click(save_background, inputs=[background, user_profile], outputs=[background_output, user_profile]) gr.Markdown("### Step 3: Start Interview") chatbot = gr.Chatbot(label="Interview Bot", type="messages") with gr.Row(): msg = gr.Textbox(label="Your message") audio_input = gr.Audio(type="filepath", label="🎙️ Upload or Record your answer") with gr.Row(): send_btn = gr.Button("Send Text") audio_btn = gr.Button("Send Audio") send_btn.click(respond, inputs=[msg, chat_history, user_profile], outputs=[chatbot], queue=False) send_btn.click(lambda: "", None, msg, queue=False) audio_btn.click(handle_audio, inputs=[audio_input, chat_history, user_profile], outputs=[chatbot], queue=False) # ✅ Step 4: Webcam Posture Check gr.Markdown("### Step 4: Webcam Posture Check") webcam = gr.Image(source="webcam", label="Capture Posture") posture_output = gr.Textbox(label="Posture Feedback") posture_btn = gr.Button("Analyze Posture") posture_btn.click(classify_image, inputs=[webcam], outputs=[posture_output]) # LAUNCH demo.launch()