import gradio as gr
import pandas as pd
import numpy as np
import faiss
import os
import re
import random
from openai import OpenAI

# Initialize OpenAI client securely
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Convert a CRM row into readable text for embeddings
def row_to_text(row):
    return (
        f"Clinic: {row.get('Account_Name', 'N/A')}\n"
        f"Interaction Focus: {row.get('Recent_Interaction_Focus', 'N/A')}\n"
        f"Interaction Notes: {row.get('Recent_Interaction_Notes', 'N/A')}\n"
        f"Region: {row.get('Region', 'Unknown')}\n"
        f"Account Manager: {row.get('Account_Manager', 'Unknown')}\n"
        f"Revenue: €{row.get('Revenue', 'N/A')}\n"
        f"Last Purchase: {row.get('Last_Purchase_Date', 'N/A')}\n"
        f"Churn Risk: {row.get('Churn_Risk', 'N/A')}\n"
        f"Satisfaction Score: {row.get('Satisfaction_Score', 'N/A')}/10\n"
        f"Product Interest: {row.get('Product_Interest', 'N/A')}\n"
        f"Notes: {row.get('Account_Notes', 'N/A')}"
    )

# Get embedding for a given text
def get_embedding(text):
    response = client.embeddings.create(
        input=[text],
        model="text-embedding-3-small"
    )
    return response.data[0].embedding

# Core function: generate contextual pitch
def contextual_pitch_assistant(csv_file, query, sender_name):
    df = pd.read_csv(csv_file.name)
    text_chunks = df.apply(row_to_text, axis=1).tolist()
    embeddings = [get_embedding(t) for t in text_chunks]

    dim = len(embeddings[0])
    index = faiss.IndexFlatL2(dim)
    index.add(np.array(embeddings).astype("float32"))

    q_emb = np.array([get_embedding(query)]).astype("float32")
    D, I = index.search(q_emb, 3)
    retrieved = [df.iloc[i] for i in I[0]]

    # Extract contextual fields
    row = retrieved[0]
    account_manager = row.get("Account_Manager", "team")
    account_name = row.get("Account_Name", "your clinic")
    focus = str(row.get("Recent_Interaction_Focus", "")).lower()
    notes_focus = str(row.get("Recent_Interaction_Notes", ""))
    notes_background = str(row.get("Account_Notes", ""))

    if not sender_name.strip():
        sender_name = "The Sales Team"

    # Map focus → style instruction
    if "business" in focus:
        style_instruction = "Emphasize ROI, efficiency, and competitive advantage."
    elif "patient" in focus:
        style_instruction = "Emphasize patient comfort, satisfaction, and outcomes."
    elif "clinical" in focus:
        style_instruction = "Emphasize precision, innovation, and clinical quality."
    else:
        style_instruction = "Balance clinical, patient, and business value."

    # Build prompt
    
    prompt = f"""
You are an expert in B2B sales messaging, inspired by Challenger Sales and Jeb Blount’s prospecting techniques. 
Your job is to write a short, natural-sounding HTML email pitch to a dental clinic.

Rules:
- Begin with "Dear {account_manager}," as the greeting.
- Anchor immediately in the clinic’s recent concern: "{notes_focus}".
- Use background context if relevant: "{notes_background}".
- Adapt the pitch style: {style_instruction}
- Apply Challenger logic: 
  * Start with the client’s concern in their own terms.
  * Reframe with an insight (show a broader problem or missed opportunity).
  * Link features → outcomes → impact (business, patient, or clinical).
  * Confident but helpful tone.
- Keep under 150 words.
- End with a strong, specific CTA (e.g. propose a short call or demo, suggest a time).
- Close with "Best regards," followed by "{sender_name}".
- Return only HTML — no markdown, no code fences.

Query:
{query}

CRM context (for your understanding, do not copy verbatim):
{row_to_text(row)}
"""


    response = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": prompt}],
        temperature=0.7
    )

    output_text = response.choices[0].message.content

    # TEMP FIX: use random pre-uploaded header image
    image_choices = [
        "https://huggingface.co/spaces/nmcamacho/RAGdemo/resolve/main/dental_header_1.png",
        "https://huggingface.co/spaces/nmcamacho/RAGdemo/resolve/main/dental_header_2.png",
        "https://huggingface.co/spaces/nmcamacho/RAGdemo/resolve/main/dental_header_3.png"
    ]
    image_url = random.choice(image_choices)

    # Clean output text
    output_text = re.sub(r"(?is)dall[-·]e.*?```", "", output_text)
    output_text = re.sub(r"###.*?HTML Email Pitch", "", output_text)
    output_text = re.sub(r"```html|```", "", output_text)
    output_text = output_text.strip()

    clinic_note = f"""
    <div style='margin-top:20px;padding:10px;font-size:13px;color:gray;
    border-top:1px solid #ddd;'>
      📌 Target: <b>{account_name}</b> (Region: {row.get('Region', 'N/A')}, 
      Focus: {row.get('Recent_Interaction_Focus', 'N/A')})
    </div>
    """

    html = f"""
    <div style='font-family:Arial,sans-serif;max-width:700px;margin:auto;padding:24px;background:#ffffff;
    border-radius:12px;box-shadow:0 3px 10px rgba(0,0,0,0.1);'>
      <img src="{image_url}" style="width:100%;border-radius:8px;margin-bottom:20px;">
      {output_text}
      {clinic_note}
    </div>
    """

    return html

# Build Gradio app (unchanged, apart from column updates)
with gr.Blocks(
    title="Contextual Pitch Assistant for Dental Sales",
    css="""
    body { background-color: #f7f9f9; font-family: 'Inter', sans-serif; }
    #output_html { min-height: 450px; }
    .gradio-container { max-width: 90% !important; margin: auto; }
    h1, h2, h3, h4, h5 { color: #00857C; font-weight: 600; }
    .gr-button { 
        background-color: #00857C !important; 
        color: white !important; 
        border: none !important;
        font-weight: 600;
        padding: 10px 18px;
        border-radius: 8px;
    }
    .gr-button:hover { background-color: #006e67 !important; }
    .gr-file, .gr-textbox {
        border: 1px solid #d1d5db !important;
        border-radius: 8px !important;
    }
    .gr-box {
        background: white !important;
        border-radius: 12px !important;
        box-shadow: 0 2px 8px rgba(0,0,0,0.05);
        padding: 20px !important;
    }
    """
) as app:
    gr.Markdown(
        """
        # 🦷 Contextual Pitch Assistant for Dental Sales  
        *Powered by contextual CRM data and generative AI*  
        ---
        Upload a CRM file and enter a sales question — get a personalized email pitch with a contextual image.
        """
    )

    with gr.Row():
        csv_file = gr.File(label="📂 Upload CRM CSV (with Recent_Interaction_Focus/Notes)", file_types=[".csv"])
        sender_name = gr.Textbox(
            label="✍️ Who signs the email?", 
            placeholder="e.g. Nuno Camacho, Sales Director", 
            value="Nuno Camacho"
        )

    query = gr.Textbox(
        label="💬 Sales Query", 
        placeholder="e.g. Which clinic is best for our imaging subscription?",
        lines=2
    )

    run_btn = gr.Button("🚀 Generate Pitch", variant="primary")
    output = gr.HTML(label="✨ Email Pitch Preview", elem_id="output_html")

    run_btn.click(fn=contextual_pitch_assistant, inputs=[csv_file, query, sender_name], outputs=output)

    run_btn.click(
        lambda: "<p style='color:gray;'>⏳ Processing... please wait 30–60 seconds.</p>",
        inputs=None,
        outputs=output,
        queue=False
    )

app.launch()