app.py

# requirements.txt
# gradio
# openai>=1.0.0
# sentence-transformers

import gradio as gr
from openai import OpenAI, AuthenticationError
from sentence_transformers import SentenceTransformer, util
import os
import time

# --- Load Resume and Build Embedding Index ---
embedder = SentenceTransformer('all-MiniLM-L6-v2')

RESUME_PATH = "resume.txt"

def load_resume_chunks(path=RESUME_PATH):
    with open(path, "r") as f:
        content = f.read()
    chunks = [chunk.strip() for chunk in content.split("\n\n") if chunk.strip()]
    embeddings = embedder.encode(chunks, convert_to_tensor=True)
    return chunks, embeddings

resume_chunks, resume_embeddings = load_resume_chunks()

def get_relevant_resume_chunks(query, top_k=3):
    query_embedding = embedder.encode(query, convert_to_tensor=True)
    scores = util.cos_sim(query_embedding, resume_embeddings)[0]
    top_results = scores.argsort(descending=True)[:top_k]
    return [resume_chunks[i] for i in top_results]

def extract_projects_section(path=RESUME_PATH):
    with open(path, "r") as f:
        content = f.read()
    start = content.find("Project:")
    end = content.find("Publication", start)
    if start != -1 and end != -1:
        return content[start:end].strip()
    return ""

# --- Audio Bot Logic ---
def audio_bot(api_key, audio_filepath, chat_history):
    if not api_key:
        chat_history.append((None, "Error: OpenAI API key is missing."))
        return chat_history, chat_history, None, None

    try:
        client = OpenAI(api_key=api_key)
    except Exception as e:
        chat_history.append((None, f"Error initializing OpenAI client: {e}"))
        return chat_history, chat_history, None, None

    if audio_filepath is None:
        chat_history.append((None, "Error: No audio recorded."))
        return chat_history, chat_history, None, None

    messages = []

    for user_msg, ai_msg in chat_history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if ai_msg:
            messages.append({"role": "assistant", "content": ai_msg})

    try:
        with open(audio_filepath, "rb") as audio_file:
            transcription = client.audio.transcriptions.create(
                model="whisper-1",
                file=audio_file
            )
        user_text = transcription.text
        print(f"User said: {user_text}")

        # Check if the user is asking about projects
        if "project" in user_text.lower():
            projects_info = extract_projects_section()
            system_prompt = f"""
            You are Thejas Haridas — a Data Scientist, ML/AI Engineer, and Data Analyst.
            Answer all questions in first person, as if you are Thejas Haridas himself.
            Only use the following projects to answer questions about your work:
            Always respond only in English.

            {projects_info}
            """
        else:
            # Get relevant resume info and similarity scores
            query_embedding = embedder.encode(user_text, convert_to_tensor=True)
            scores = util.cos_sim(query_embedding, resume_embeddings)[0]
            top_results = scores.argsort(descending=True)[:5]  # Increase top_k to 5
            relevant_info = "\n\n".join([resume_chunks[i] for i in top_results])
            max_score = float(scores[top_results[0]])
            system_prompt = f"""
            You are Thejas Haridas — a Data Scientist, ML/AI Engineer, and Data Analyst.
            Answer all questions in first person, as if you are Thejas Haridas himself.
            Use the following relevant background info to answer questions:
            Always respond only in English.

            {relevant_info}
            """

        messages.insert(0, {"role": "system", "content": system_prompt})
        messages.append({"role": "user", "content": user_text})

        response = client.chat.completions.create(
            model="gpt-4o-mini",
            messages=messages
        )
        ai_text = response.choices[0].message.content
        print(f"AI responded: {ai_text}")

        chat_history.append((user_text, ai_text))

        # Convert response to speech
        output_audio_filepath = f"ai_response_{int(time.time())}.mp3"
        speech_response = client.audio.speech.create(
            model="tts-1",
            voice="nova",
            input=ai_text
        )
        speech_response.stream_to_file(output_audio_filepath)

        return chat_history, chat_history, output_audio_filepath, None

    except AuthenticationError:
        error_msg = "Error: Invalid OpenAI API key."
        chat_history.append((user_text if 'user_text' in locals() else "Audio input", error_msg))
        return chat_history, chat_history, None, None
    except Exception as e:
        error_msg = f"An error occurred: {e}"
        print(error_msg)
        chat_history.append((user_text if 'user_text' in locals() else "Audio input", error_msg))
        return chat_history, chat_history, None, None
    finally:
        if audio_filepath and os.path.exists(audio_filepath):
            os.remove(audio_filepath)

# --- Gradio UI ---
with gr.Blocks(theme=gr.themes.Soft(), title="Conversational Resume Bot") as demo:
    gr.Markdown("# 🤖 Talk to Thejas Haridas (AI Resume Bot)")
    gr.Markdown("Use your voice to ask questions about Thejas Haridas' background and experience.")

    chat_history_state = gr.State([])

    with gr.Row():
        api_key_input = gr.Textbox(
            label="OpenAI API Key",
            placeholder="Enter your OpenAI API key here...",
            type="password"
        )
    
    with gr.Row():
        gr.Markdown("⚠️ **After recording, please click the ▶️ play button once before submitting, or the audio won't be saved.**")

    
    with gr.Row():
        audio_input = gr.Audio(
            sources=["microphone"],
            type="filepath",
            label="Record your message",
            streaming=False,  # <-- Add this line to ensure file is saved on submit
            interactive=True
        )

    with gr.Row():
        submit_button = gr.Button("Submit Audio")

    with gr.Column():
        chatbot_output = gr.Chatbot(label="Conversation", height=400, avatar_images=("./user_avatar.png", "./bot_avatar.png"))
        audio_output = gr.Audio(label="AI Response", autoplay=True, interactive=False, format="mp3")

    submit_button.click(
        fn=audio_bot,
        inputs=[api_key_input, audio_input, chat_history_state],
        outputs=[chatbot_output, chat_history_state, audio_output, audio_input]  # Add audio_input here
    )

    gr.Markdown("---")
    gr.Markdown("Made with ❤️ by Thejas Haridas using Gradio and OpenAI.")

if __name__ == "__main__":
    if not os.path.exists("user_avatar.png"):
        with open("user_avatar.png", "w"): pass
    if not os.path.exists("bot_avatar.png"):
        with open("bot_avatar.png", "w"): pass
    demo.launch(debug=True, share=True)