Update app.py

app.py

@@ -1,71 +1,37 @@
 import pandas as pd
-import numpy as np
-import re
-from transformers import pipeline
 import gradio as gr
+from transformers import pipeline
+from langchain.llms import HuggingFacePipeline
+from langchain.agents import create_pandas_dataframe_agent
 
-# Load numeric data
+# Load your data
 df = pd.read_csv("synthetic_profit.csv")
-df_str = df.astype(str)
 
-# Initialize TAPAS
-qa = pipeline(
-    "table-question-answering",
-    model="google/tapas-base-finetuned-sqa",
-    tokenizer="google/tapas-base-finetuned-sqa"
+# Set up lightweight Hugging Face pipeline with Flan-T5 base
+hf_pipeline = pipeline(
+    task="text2text-generation",
+    model="google/flan-t5-base",
+    device=-1  # ensures CPU
 )
 
-# Ensure column names are properly capitalized
-metric_columns = {"profitmargin": "ProfitMargin", "profit": "Profit", 
-                  "revenue": "Revenue", "cost": "Cost"}
-
-def answer(q: str) -> str:
-    # 1. Conditional query: negative profit
-    if re.search(r"products.*negative.*profit", q, re.IGNORECASE):
-        negative_profits = df[df["Profit"] < 0]
-        if negative_profits.empty:
-            return "✅ No products with negative profit found."
-        results = negative_profits[['Product', 'Region', 'FiscalQuarter', 'FiscalYear', 'Profit']]
-        return results.to_string(index=False)
-
-    # 2. Numeric summaries (total/average) – Case insensitive metric matching
-    match = re.search(
-        r"\b(total|average)\s+(profitmargin|profit|revenue|cost)\b.*\bProduct\s*([A-D])\b.*\b(Q[1-4])\s+(\d{4})",
-        q, re.IGNORECASE
-    )
-    if match:
-        agg_type, metric_key, product_letter, quarter, year = match.groups()
-        metric = metric_columns[metric_key.lower()]
-        product = f"Product {product_letter.upper()}"
-        subset = df[
-            (df["Product"] == product) &
-            (df["FiscalQuarter"] == quarter) &
-            (df["FiscalYear"] == int(year))
-        ]
-        if subset.empty:
-            return "⚠️ No matching data found."
-        if agg_type.lower() == "total":
-            val = subset[metric].sum()
-        else:  # average
-            val = subset[metric].mean()
-
-        formatted_val = f"{val:.3f}" if metric == "ProfitMargin" else f"{val:,.2f}"
-        return f"📊 {agg_type.title()} {metric} for {product} in {quarter} {year}: {formatted_val}"
+# LangChain agent setup
+llm = HuggingFacePipeline(pipeline=hf_pipeline)
+agent = create_pandas_dataframe_agent(llm, df, verbose=True)
 
-    # 3. TAPAS fallback (with error handling)
+def answer(query: str) -> str:
     try:
-        res = qa(table=df_str, query=q)
-        return f"🔍 {res['answer']} (agg: {res.get('aggregator','NONE')})"
+        response = agent.run(query)
+        return f"📊 {response}"
     except Exception as e:
-        return f"❗ Error from TAPAS: {str(e)}"
+        return f"❗ Error: {str(e)}"
 
-# Launch Gradio
+# Gradio interface
 demo = gr.Interface(
     fn=answer,
-    inputs=gr.Textbox(lines=2, placeholder="e.g. 'total Revenue for Product B in Q1 2024?' or 'List products with negative profit.'"),
+    inputs=gr.Textbox(lines=2, placeholder="E.g., 'Total revenue for Product B in EMEA', 'List products with negative profit.'"),
     outputs="text",
-    title="🟢 SAP S/4HANA Profitability Chat",
-    description="Ask questions on profitability data (synthetic demo). Supports total, average, and conditional queries."
+    title="🟢 SAP Profitability Data Chat (Flan-T5 + Pandas)",
+    description="Ask questions about synthetic SAP profitability data. Powered by Flan-T5-base via Hugging Face."
 )
 
 demo.launch()