# app.py
import gradio as gr
from huggingface_hub import InferenceClient
import os
from dotenv import load_dotenv

# Load environment variables from .env file for local development
load_dotenv()

# Set up the Hugging Face Client using the API Key
HF_TOKEN = os.getenv("HUGGINGFACE_API_KEY")
client = InferenceClient(token=HF_TOKEN)

def analyze_feedback(user_feedback):
    """
    This function takes user feedback, sends it to the HuggingFace Inference API,
    and returns a structured analysis.
    """
    # This high-quality prompt is well-suited for an instruction-tuned model.
    prompt = f"""
    You are a world-class Senior Product Manager, an expert in qualitative data analysis.
    Your mission is to analyze the following user feedback text.
    Provide a response structured in Markdown format. The response must include:

    1.  **Executive Summary (max 3 sentences):** The main idea that emerges.
    2.  **Key Positive Themes (3 points):** The most appreciated aspects, with a short quote for each.
    3.  **Key Negative Themes / Friction Points (3 points):** The most recurring problems, with a short quote for each.
    4.  **Actionable Recommendations (2 suggestions):** Propose two concrete actions the product team could consider.

    ---
    User Feedback to Analyze:
    {user_feedback}
    """

    # We must format the prompt into the message structure required by the 'conversational' task.
    messages = [{"role": "user", "content": prompt}]

    # We use a try...except block to gracefully handle potential API errors.
    try:
        # Calling the chat_completion endpoint as required by this model's provider.
        response = client.chat_completion(
            messages=messages,
            model="mistralai/Mistral-7B-Instruct-v0.2",
            max_tokens=1024, # Renamed from max_new_tokens for this endpoint
        )
        
        # The response from chat_completion is a structured object, so we extract the content.
        return response.choices[0].message.content
        
    except Exception as e:
        return f"An error occurred: {e}"

# Create the Gradio interface
iface = gr.Interface(
    fn=analyze_feedback,
    inputs=gr.Textbox(lines=15, placeholder="Paste your raw user feedback here..."),
    outputs=gr.Markdown(),
    title="💡 Insight Synthesizer",
    description="An AI-powered tool for Product Managers to quickly synthesize raw user feedback into actionable insights. This is an MVP built for a portfolio project.",
    theme=gr.themes.Soft(),
    allow_flagging="never"
)

# Launch the app
iface.launch()