app.py

"""
Gini AI - Full-stack web development assistant
Gradio version of the original JS-Coder-Backend (Node.js + @huggingface/transformers)
Created by Emalawi19. Founded by Professor Roosevelt Chinkwende.
"""

import threading
import time
from datetime import datetime

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

# ── CONFIG ───────────────────────────────────────────────────────────────────
MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"
LOG_FILE = "generation_logs.txt"

# ── LOGGING ──────────────────────────────────────────────────────────────────
def log(event: str, details: dict | None = None):
    details = details or {}
    timestamp = datetime.utcnow().isoformat()
    entry = f"[{timestamp}] {event.upper()} - {details}\n"
    print(entry, end="")
    with open(LOG_FILE, "a", encoding="utf-8") as f:
        f.write(entry)


# ── MODEL LOADING ────────────────────────────────────────────────────────────
log("system", {"message": "Loading model..."})

device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
).to(device)
model.eval()

log("system", {"message": "Model ready!", "device": device})


# ── PROMPT CLASSIFIER ────────────────────────────────────────────────────────
def classify_prompt(prompt: str) -> str:
    lower = prompt.lower()

    identity_words = [
        "who are you", "what are you", "your name", "who made you",
        "who created you", "your company", "your founder", "about you",
        "introduce yourself",
    ]
    if any(w in lower for w in identity_words):
        return "identity"

    general_words = [
        "what is", "what are", "explain", "describe", "tell me about",
        "history of", "meaning of", "definition", "how does", "why is",
        "who is", "who was", "when did", "where is", "function of",
        "functions of",
    ]
    if any(w in lower for w in general_words):
        return "general"

    coding_words = [
        "generate", "create", "build", "make", "write", "code", "website",
        "webpage", "page", "app", "application", "login", "register", "form",
        "html", "css", "javascript", "js", "node", "express", "backend",
        "frontend", "server", "api", "function", "script", "component",
        "template", "dashboard", "portfolio", "ecommerce", "shop", "blog",
        "landing page", "navbar", "footer", "button", "database",
    ]
    if any(w in lower for w in coding_words):
        return "coding"

    return "general"


# ── SYSTEM PROMPT BUILDER ────────────────────────────────────────────────────
IDENTITY = """You are Gini AI, a full-stack web development AI assistant.
- Created by Emalawi19.
- Founded by Professor Roosevelt Chinkwende.
- You are an AI assistant, not a human.
- When asked your name: "I am Gini AI."
- When asked who made you: "I was created by Emalawi19."
- When asked about the founder: "Emalawi19 was founded by Professor Roosevelt Chinkwende."."""


def build_system_prompt(prompt_type: str) -> str:
    if prompt_type == "identity":
        return f"""{IDENTITY}
Introduce yourself clearly and warmly. State your name, your purpose, your company, and your founder."""

    if prompt_type == "general":
        return f"""{IDENTITY}
GENERAL BEHAVIOR:
- Answer the question directly and clearly in plain English.
- Be concise and helpful.
- Do NOT generate any code for non-coding questions.
- Do NOT use programming languages to answer everyday questions."""

    # prompt_type == "coding"
    return f"""{IDENTITY}

YOU ARE A FULL-STACK WEB DEVELOPER AI. FOLLOW THESE RULES STRICTLY:

RULE 1 — SINGLE FILE OUTPUT (MOST IMPORTANT):
When asked to build any website, webpage, app, or UI component:
- ALWAYS output ONE single complete HTML file.
- That single file MUST contain ALL HTML structure, ALL CSS styles, and ALL JavaScript — nothing external.
- ALL CSS goes inside a <style> tag inside <head>.
- ALL JavaScript goes inside a <script> tag at the bottom of <body>.
- NEVER say "create a separate CSS file" or "create a separate JS file".
- NEVER split code across multiple files unless the user specifically asks for backend server code.

RULE 2 — COMPLETE CODE ONLY:
- The file must be 100% complete and copy-paste ready.
- Start ALWAYS with <!DOCTYPE html> on the very first line.
- End ALWAYS with </body> then </html> as the absolute last lines.
- NEVER use placeholders like "add your code here" or "// TODO".
- NEVER truncate or cut off. Always finish the complete file.

RULE 3 — CODE QUALITY:
- Write clean, modern, well-commented HTML/CSS/JS.
- Use responsive design (flexbox or grid, mobile-friendly).
- Include hover effects, smooth transitions, and professional styling.
- Use CSS custom properties (variables) for colors and theming.
- JavaScript must be functional — forms should validate, buttons should work.

RULE 4 — BACKEND REQUESTS:
- If the user asks for a backend (Node.js, Express, API, server), provide it as a SEPARATE clearly labeled code block AFTER the frontend file.
- Backend code goes in a ```javascript block labeled "server.js".
- Still provide the complete frontend HTML file first.

RULE 5 — STEP BY STEP (only for complex multi-feature apps):
- If building a complex app, give a brief overview of what the file contains BEFORE the code.
- Then provide the single complete file.
- End with: "This is the complete file. Copy and save it as index.html and open in your browser."

RULE 6 — FORMAT:
- Always wrap the HTML file in a ```html code block.
- Always wrap any JS server code in a ```javascript code block.
- Add a short explanation after the code of what was built and how to use it."""


# ── TOKEN BUDGET PER TYPE ────────────────────────────────────────────────────
def token_budget(prompt_type: str) -> int:
    return {"identity": 200, "general": 500, "coding": 2000}.get(prompt_type, 500)


# ── STREAMING GENERATION ─────────────────────────────────────────────────────
def respond(message: str, history: list[dict]):
    """
    Gradio ChatInterface streaming handler (type="messages").
    history is a list of {"role": ..., "content": ...} dicts of prior turns.
    """
    if not message or not message.strip():
        yield "Please enter a prompt."
        return

    prompt_type = classify_prompt(message)
    system_prompt = build_system_prompt(prompt_type)
    max_new_tokens = token_budget(prompt_type)

    log("job_started", {
        "promptType": prompt_type,
        "maxTokens": max_new_tokens,
        "prompt": message[:80],
    })

    # Build the chat messages: system + prior turns + new user message
    messages = [{"role": "system", "content": system_prompt}]
    for turn in history:
        if turn.get("role") in ("user", "assistant") and turn.get("content"):
            messages.append({"role": turn["role"], "content": turn["content"]})
    messages.append({"role": "user", "content": message})

    input_ids = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True, return_tensors="pt"
    ).to(device)

    streamer = TextIteratorStreamer(
        tokenizer, skip_prompt=True, skip_special_tokens=True
    )

    generation_kwargs = dict(
        input_ids=input_ids,
        max_new_tokens=max_new_tokens,
        temperature=0.2,
        do_sample=False,
        repetition_penalty=1.15,
        streamer=streamer,
    )

    start_time = time.time()
    thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    partial_text = ""
    for new_text in streamer:
        partial_text += new_text
        yield partial_text

    thread.join()

    # Repair unclosed code blocks, same as the original Node.js chunk repair
    open_blocks = partial_text.count("```")
    if open_blocks % 2 != 0:
        partial_text += "\n```"
        yield partial_text

    duration = time.time() - start_time
    log("job_completed", {
        "promptType": prompt_type,
        "durationSec": round(duration, 2),
        "outputChars": len(partial_text),
    })


# ── GRADIO UI ─────────────────────────────────────────────────────────────────
demo = gr.ChatInterface(
    fn=respond,
    type="messages",
    title="Gini AI — Full-Stack Web Development Assistant",
    description=(
        "Built by **Emalawi19** · Founded by **Professor Roosevelt Chinkwende**. "
        "Ask Gini AI to build a website, explain a concept, or just say hi."
    ),
    examples=[
        "Who are you?",
        "Build a responsive login page with a gradient background",
        "Explain how a REST API works",
    ],
    theme=gr.themes.Soft(primary_hue="purple"),
)

if __name__ == "__main__":
    demo.queue().launch()