app.py

import gradio as gr
import pandas as pd
from transformers import pipeline

# 1) Load & stringify your CSV
df = pd.read_csv("synthetic_profit.csv")
table = df.astype(str).to_dict(orient="records")

# 2) Instantiate the TAPAS pipeline from Transformers
qa = pipeline(
    "table-question-answering",
    model="google/tapas-base-finetuned-wtq",
    tokenizer="google/tapas-base-finetuned-wtq",
    device=-1,   # CPU; change to 0 if you have a GPU
)

# 3) Few-shot examples teach “filter + sum” vs. “filter + mean”
EXAMPLES = """
Example 1:
Q: What is the total revenue for Product A in EMEA in Q1 2024?
A: Filter Product=A & Region=EMEA & FiscalYear=2024 & FiscalQuarter=Q1, then sum Revenue → 3075162.49

Example 2:
Q: What is the total cost for Product A in EMEA in Q1 2024?
A: Filter Product=A & Region=EMEA & FiscalYear=2024 & FiscalQuarter=Q1, then sum Cost → 2894321.75

Example 3:
Q: What is the total margin for Product A in EMEA in Q1 2024?
A: Filter Product=A & Region=EMEA & FiscalYear=2024 & FiscalQuarter=Q1, then sum ProfitMargin → 0.18

Example 4:
Q: What is the average profit margin for Product A in EMEA in Q1 2024?
A: Filter Product=A & Region=EMEA & FiscalYear=2024 & FiscalQuarter=Q1, then mean ProfitMargin → 0.18
"""

def answer_question(question: str) -> str:
    prompt = EXAMPLES + f"\nQ: {question}\nA:"
    try:
        result = qa(table=table, query=prompt)
        return result.get("answer", "No answer found.")
    except Exception as e:
        return f"❌ Pipeline error:\n{e}"

# 4) Gradio UI
iface = gr.Interface(
    fn=answer_question,
    inputs=gr.Textbox(lines=2, placeholder="e.g. What is the total revenue for Product A in Q1 2024?"),
    outputs=gr.Textbox(lines=3),
    title="SAP Profitability Q&A",
    description=(
        "Ask simple sum/mean questions on the synthetic SAP data.  \n"
        "Powered by google/tapas-base-finetuned-wtq with four few-shot examples."
    ),
    allow_flagging="never",
)

if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)