app.py

import gradio as gr
from inference import predict
from logger import log_feedback_to_firebase
import openai
import os
from watchdog import manual_retrain
from dashboard import render_dashboard

openai.api_key = os.environ["OPENAI_API_KEY"]  # Add this secret in your Hugging Face Space

def gpt_predict(goal, sol1, sol2):
    prompt = f"You're solving a commonsense reasoning problem. Choose the better solution to achieve the goal.\n\nGoal: {goal}\n\nOption 1: {sol1}\nOption 2: {sol2}\n\nWhich option is better? Reply with only 'Solution 1' or 'Solution 2'."
    try:
        response = openai.ChatCompletion.create(
            model="gpt-3.5-turbo",
            messages=[{"role": "user", "content": prompt}],
            temperature=0
        )
        reply = response.choices[0].message["content"]
        return "Solution 1" if "1" in reply else "Solution 2"
    except Exception as e:
        return f"GPT Error: {str(e)}"

def compare(goal, sol1, sol2, correct):
    evo = predict(goal, sol1, sol2)
    gpt = gpt_predict(goal, sol1, sol2)

    if evo == gpt == correct:
        verdict = "✅ Both Evo and GPT-3.5 are correct!"
    elif evo == correct and gpt != correct:
        verdict = "🧠 Evo got it right. GPT-3.5 missed it."
    elif gpt == correct and evo != correct:
        verdict = "🤖 GPT-3.5 got it right. Evo missed it."
    else:
        verdict = "❌ Both models got it wrong."

    log_feedback_to_firebase(goal, sol1, sol2, evo, gpt, correct, "from app.py")

    return evo, gpt, verdict

def trigger_retrain():
    success = manual_retrain()
    return "✅ Evo retrained successfully!" if success else "⚠️ Retraining failed."

with gr.Blocks() as demo:
    with gr.Tab("⚔️ Evo vs GPT-3.5 Showdown"):
        gr.Markdown("## ⚔️ Evo vs GPT-3.5 – Real-Time Commonsense Showdown")
        gr.Markdown(
            "> 🧠 EvoTransformer v2.1 – PIQA Accuracy: 69.7% (vs GPT-3.5 ≈ 81%)\n"
            "> 13M Parameters • Fully Scratch-Trained • Leans Smart\n"
            "> 🧪 *Note: EvoTransformer is a scratch-built model trained on 1K PIQA examples. It may occasionally misinterpret context or idioms. That’s part of its evolution.*"
        )
        gr.Markdown(
            "> 🔄 *EvoTransformer is not static. Every time you provide feedback, Evo learns and evolves. Welcome to real-time neural evolution.*"
        )

        with gr.Row():
            goal = gr.Text(label="Goal")
        with gr.Row():
            sol1 = gr.Text(label="Solution 1")
            sol2 = gr.Text(label="Solution 2")
        correct = gr.Radio(choices=["Solution 1", "Solution 2"], label="✅ Which is actually correct?", value="Solution 1")

        btn = gr.Button("Submit")
        evo_out = gr.Text(label="🧠 EvoTransformer Response")
        gpt_out = gr.Text(label="🤖 GPT-3.5 Response")
        verdict_out = gr.Text(label="⚖️ Verdict")

        btn.click(fn=compare, inputs=[goal, sol1, sol2, correct], outputs=[evo_out, gpt_out, verdict_out])

        gr.Markdown("#### 🔍 Try These Examples:")
        examples = [
            ["Start a fire", "Use a match", "Pour water", "Solution 1"],
            ["Warm up food", "Use microwave", "Put it in fridge", "Solution 1"],
            ["Charge a phone", "Plug it in", "Put it on grass", "Solution 1"],
            ["Get rid of bad smell", "Open window", "Close door", "Solution 1"],
        ]
        gr.Examples(examples=examples, inputs=[goal, sol1, sol2, correct])

        retrain_btn = gr.Button("🔁 Retrain EvoTransformer")
        retrain_status = gr.Text(label="📢 Retrain Status")
        retrain_btn.click(fn=trigger_retrain, outputs=[retrain_status])

        gr.Markdown("Made with ❤️ by Dr. Heman Mohabeer — EvoTransformer is not just code. It's evolution.")

    with gr.Tab("📊 Evo Dashboard"):
        render_dashboard()

demo.launch()