app.py

import json
import pandas as pd
import matplotlib.pyplot as plt
import gradio as gr
from openai import OpenAI

# ==================================================
# OpenAI client helper (API key from user)
# ==================================================
def get_openai_client(api_key: str):
    return OpenAI(api_key=api_key)

# ==================================================
# CSV Loader + Normalizer
# ==================================================
def load_and_normalize_csv(csv_file):
    df = pd.read_csv(csv_file)

    # Normalize column names
    df.columns = (
        df.columns
        .str.strip()
        .str.lower()
        .str.replace(" ", "")
        .str.replace("/", "")
        .str.replace("_", "")
    )

    # Normalize drcr column if present
    if "drcr" in df.columns:
        df["drcr"] = (
            df["drcr"]
            .astype(str)
            .str.strip()
            .str.lower()
            .replace({
                "cr": "credit",
                "db": "debit"
            })
        )

    return df

# ==================================================
# Ask OpenAI for INTENT (STRICT JSON)
# ==================================================
def get_intent(question: str, api_key: str) -> dict:
    client = get_openai_client(api_key)

    response = client.chat.completions.create(
        model="gpt-4o-mini",
        response_format={"type": "json_object"},
        messages=[
            {
                "role": "system",
                "content": (
                    "You are a data analysis planner.\n"
                    "Return ONLY valid JSON.\n"
                    "Do NOT explain.\n\n"
                    "JSON format:\n"
                    "{\n"
                    '  "action": "count | sum | plot",\n'
                    '  "filters": { "year": number | null, "drcr": string | null },\n'
                    '  "groupby": "year | drcr | null"\n'
                    "}"
                )
            },
            {
                "role": "user",
                "content": question
            }
        ]
    )

    return json.loads(response.choices[0].message.content)

# ==================================================
# Execute intent using real Python
# ==================================================
def execute_intent(intent: dict, df: pd.DataFrame):
    data = df.copy()

    # Apply filters
    filters = intent.get("filters", {})
    for key, value in filters.items():
        if value is not None and key in data.columns:
            data = data[data[key] == value]

    action = intent.get("action")
    group_col = intent.get("groupby")

    # COUNT
    if action == "count":
        if group_col:
            return data.groupby(group_col).size()
        return len(data)

    # SUM
    if action == "sum":
        if group_col:
            return data.groupby(group_col)["amount"].sum()
        return data["amount"].sum()

    # PLOT
    if action == "plot":
        if not group_col:
            raise ValueError("Plot requires groupby")

        result = data.groupby(group_col).size()
        result.plot(kind="bar")
        plt.title("Result")
        plt.tight_layout()
        plt.show()
        return result

    raise ValueError(f"Unknown action: {action}")

# ==================================================
# End-to-end question answering
# ==================================================
def answer_question(question: str, api_key: str, df: pd.DataFrame):
    intent = get_intent(question, api_key)
    return execute_intent(intent, df)

# ==================================================
# Gradio wrapper
# ==================================================
def gradio_answer(api_key, csv_file, question):
    try:
        if not api_key:
            return "Please provide your OpenAI API key."

        if csv_file is None:
            return "Please upload a CSV file."

        if not question:
            return "Please enter a question."

        df = load_and_normalize_csv(csv_file)
        result = answer_question(question, api_key, df)

        if hasattr(result, "to_string"):
            return result.to_string()

        return str(result)

    except Exception as e:
        return f"Error: {str(e)}"

# ==================================================
# Gradio Interface (Spaces entry point)
# ==================================================
demo = gr.Interface(
    fn=gradio_answer,
    inputs=[
        gr.Textbox(label="OpenAI API Key", type="password"),
        gr.File(label="Upload CSV File", file_types=[".csv"]),
        gr.Textbox(
            label="Ask a question about your CSV",
            placeholder="How many credit operations happened in 2022?"
        )
    ],
    outputs=gr.Textbox(label="Answer"),
    title="Chat with your CSV 📊",
    description="Upload any CSV file and ask natural language questions about it"
)

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