app.py

import gradio as gr
import os
import logging

# ---------------------------------------------------------------------------
# Logging
# ---------------------------------------------------------------------------
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s [%(levelname)s] %(message)s",
    datefmt="%Y-%m-%d %H:%M:%S",
)
logger = logging.getLogger("SpoutSpoon")

# ---------------------------------------------------------------------------
# Configuration
# ---------------------------------------------------------------------------
HF_MODEL = "Qwen/Qwen2.5-Coder-3B-Instruct:nscale"
HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")

# ---------------------------------------------------------------------------
# System Prompt
# ---------------------------------------------------------------------------
SYSTEM_PROMPT = """You are Sprout & Spoon, a concise and helpful assistant for cooking and gardening advice.

Rules you MUST follow:
- Do NOT include any conversational filler. No greetings, no 'Hello', no 'Hope this helps', no 'Let me know if...'.
- Use strict Markdown formatting with **bold headers** and bullet points where appropriate.
- Keep answers short, direct, and easy to read.
- Use large, easy-to-read text structure (short paragraphs, clear separation)."""


# ---------------------------------------------------------------------------
# Real LLM call via Hugging Face InferenceClient
# ---------------------------------------------------------------------------
def call_local_model(prompt: str) -> str:
    prompt_preview = prompt.strip()[:60].replace("\n", " ")
    logger.info("Received question: \"%s\"", prompt_preview)

    if not HF_API_TOKEN:
        logger.warning("HF_API_TOKEN not set - using fallback responses")
        return _fallback_response(prompt)

    logger.info(
        "Sending request to Hugging Face Inference API (model=%s)", HF_MODEL
    )

    try:
        from huggingface_hub import InferenceClient

        client = InferenceClient(token=HF_API_TOKEN)

        messages = [
            {"role": "system", "content": SYSTEM_PROMPT},
            {"role": "user", "content": prompt},
        ]

        stream = client.chat.completions.create(
            model=HF_MODEL,
            messages=messages,
            max_tokens=512,
            temperature=0.3,
            top_p=0.9,
            stream=False,
        )

        answer = stream.choices[0].message.content.strip()
        logger.info(
            "API call succeeded (response_length=%s chars)", len(answer)
        )
        return answer

    except ImportError:
        logger.warning(
            "huggingface_hub not installed - falling back to keyword response"
        )
        return _fallback_response(prompt)
    except Exception as exc:
        logger.error(
            "API request failed: %s - falling back to keyword response", exc
        )
        return _fallback_response(prompt)


# ---------------------------------------------------------------------------
# Fallback responses
# ---------------------------------------------------------------------------
def _fallback_response(prompt: str) -> str:
    lower = prompt.lower()

    if "tomato" in lower or "tomatoes" in lower:
        return (
            "**Watering**\n"
            "- Water deeply 2-3 times per week, early in the morning.\n"
            "- Avoid wetting the leaves to prevent blight.\n\n"
            "**Feeding**\n"
            "- Apply a balanced 10-10-10 fertiliser every 2 weeks.\n\n"
            "**Support**\n"
            "- Use stakes or cages once the plant is 12 inches tall.\n"
            "- Tie main stem loosely with soft garden twine."
        )

    if "chicken" in lower or "leftover" in lower:
        return (
            "**Quick Chicken Salad**\n"
            "- Shred leftover chicken and mix with Greek yoghurt, diced celery, "
            "grapes, and a pinch of salt.\n\n"
            "**Chicken and Veggie Stir-Fry**\n"
            "- Slice chicken, stir-fry with broccoli, bell peppers, and soy "
            "sauce for 5 minutes.\n\n"
            "**Warming Soup**\n"
            "- Simmer chicken with broth, carrots, onions, and egg noodles "
            "for 20 minutes."
        )

    if "rose" in lower or "prune" in lower:
        return (
            "**When to Prune**\n"
            "- Late winter or early spring, just before new growth begins.\n\n"
            "**How to Prune**\n"
            "- Remove dead, damaged, or crossing branches first.\n"
            "- Cut at a 45 degree angle 1/4 inch above an outward-facing bud.\n"
            "- Open the centre of the plant for airflow.\n\n"
            "**Aftercare**\n"
            "- Apply a layer of mulch and water thoroughly."
        )

    return (
        "**Quick Tips**\n"
        "- Keep your workspace clean and organised.\n"
        "- Prep all ingredients before you start cooking.\n"
        "- In the garden, water deeply and less often for stronger roots."
    )


# ---------------------------------------------------------------------------
# Gradio application
# ---------------------------------------------------------------------------
CUSTOM_CSS = """
.gradio-container { max-width: 800px; margin: auto; }
label { font-size: 1.2rem !important; }
button { font-size: 1.1rem !important; }
.md_output p, .md_output li { font-size: 1.4rem !important; line-height: 1.6; }
"""

with gr.Blocks(title="Sprout & Spoon") as demo:

    gr.Markdown("# \U0001f373 Sprout & Spoon\nAsk a cooking or gardening question below.")

    user_input = gr.Textbox(
        label="Your question",
        placeholder="e.g. How do I store fresh basil?",
        lines=3,
    )

    with gr.Row():
        submit_btn = gr.Button("Submit", variant="primary")
        clear_btn = gr.Button("Clear")

    # Native Markdown output — no manual HTML conversion
    output = gr.Markdown(
        value="_No answer yet._",
        label="Answer",
        elem_classes="md_output",
    )

    # Hidden textarea holding the raw markdown for the copy button
    raw_holder = gr.Textbox(
        value="",
        label="",
        visible=False,
        elem_id="raw-text-holder",
    )

    gr.Markdown("### Try an example")
    with gr.Row():
        example_tomato = gr.Button("\U0001f345 Help with my Tomatoes")
        example_chicken = gr.Button("\U0001f357 Leftover Chicken Recipe")
        example_rose = gr.Button("\U0001f339 How to prune Roses")

    def respond(message: str):
        if not message or not message.strip():
            empty = "_Please enter a question._"
            return empty, empty
        answer = call_local_model(message)
        return answer, answer

    submit_btn.click(
        fn=respond, inputs=user_input, outputs=[output, raw_holder]
    )
    user_input.submit(
        fn=respond, inputs=user_input, outputs=[output, raw_holder]
    )

    def clear_all():
        return "", "_No answer yet._", ""

    clear_btn.click(
        fn=clear_all,
        inputs=[],
        outputs=[user_input, output, raw_holder],
    )

    for btn, text in [
        (example_tomato, "Help with my Tomatoes"),
        (example_chicken, "Leftover Chicken Recipe"),
        (example_rose, "How to prune Roses"),
    ]:
        btn.click(
            fn=lambda q=text: q,
            inputs=[],
            outputs=user_input,
        ).then(
            fn=respond,
            inputs=user_input,
            outputs=[output, raw_holder],
        )


if __name__ == "__main__":
    demo.launch(share=True, theme=gr.themes.Soft(), css=CUSTOM_CSS)