Update app.py

app.py

@@ -4,29 +4,27 @@ import pandas as pd
 import duckdb
 import openai
 
-# 1) load your API key from the HF Secret
+# 1) Read your OpenAI key from the Space’s Secrets
 openai.api_key = os.getenv("OPENAI_API_KEY")
 
-# 2) load data & register in DuckDB
+# 2) Load your synthetic data into DuckDB
 df = pd.read_csv('synthetic_profit.csv')
 conn = duckdb.connect(':memory:')
 conn.register('sap', df)
 
-# ─── 3) One-line schema description for prompting ─────────────────────────────
-schema = ", ".join(df.columns)  
-# e.g. "Region,Product,FiscalYear,FiscalQuarter,Revenue,Profit,ProfitMargin"
+# 3) Build a one-line schema description for prompts
+schema = ", ".join(df.columns)  # e.g. "Region,Product,FiscalYear,FiscalQuarter,Revenue,Profit,ProfitMargin"
 
-# ─── 4) Function to call OpenAI and generate SQL ──────────────────────────────
+# 4) SQL-generation via OpenAI
 def generate_sql(question: str) -> str:
     system = (
         f"You are an expert SQL generator for a DuckDB table named `sap` "
         f"with columns: {schema}. "
-        "Generate a valid SQL query that returns exactly what the user is asking. "
-        "Only return the SQL query, without any explanation."
+        "Translate the user’s question into a valid SQL query and return _only_ the SQL."
     )
     messages = [
-        {"role": "system",  "content": system},
-        {"role": "user",    "content": question}
+        {"role": "system", "content": system},
+        {"role": "user",   "content": question},
     ]
     resp = openai.ChatCompletion.create(
         model="gpt-3.5-turbo",
@@ -35,40 +33,38 @@ def generate_sql(question: str) -> str:
         max_tokens=150,
     )
     sql = resp.choices[0].message.content.strip()
-    # Strip triple-backticks if present
-    if sql.startswith("```") and "```" in sql[3:]:
+    # strip triple-backticks if present
+    if sql.startswith("```") and sql.endswith("```"):
         sql = "\n".join(sql.splitlines()[1:-1])
     return sql
 
-# ─── 5) Core QA function: NL → SQL → execute → format result ─────────────────
+# 5) Core QA function: NL → SQL → execute → format
 def answer_profitability(question: str) -> str:
-    # 5a) Generate SQL
+    # a) generate SQL
     sql = generate_sql(question)
-
-    # 5b) Try to run it
+    # b) try to run it
     try:
-        out_df = conn.execute(sql).df()
+        result_df = conn.execute(sql).df()
     except Exception as e:
         return (
             f"❌ **Error executing SQL**\n\n"
             f"```\n{e}\n```\n\n"
             f"**Generated SQL**\n```sql\n{sql}\n```"
         )
+    # c) format the result
+    if result_df.empty:
+        return f"No rows returned.\n\n**Generated SQL**\n```sql\n{sql}\n```"
+    # single-cell → just the value
+    if result_df.shape == (1,1):
+        return str(result_df.iat[0,0])
+    # otherwise, markdown table
+    return result_df.to_markdown(index=False)
 
-    # 5c) Format the successful result
-    if out_df.empty:
-        return f"No rows returned.\n\n**SQL**\n```sql\n{sql}\n```"
-    # Single‐cell result → scalar
-    if out_df.shape == (1,1):
-        return str(out_df.iat[0,0])
-    # Otherwise → markdown table
-    return out_df.to_markdown(index=False)
-
-# ─── 6) Gradio UI ─────────────────────────────────────────────────────────────
+# 6) Gradio UI
 iface = gr.Interface(
     fn=answer_profitability,
     inputs=gr.Textbox(lines=2, placeholder="Ask a question about profitability…"),
-    outputs=gr.Markdown(),
+    outputs=gr.Markdown(),             # renders errors, code, and tables nicely
     title="SAP Profitability Q&A (OpenAI → SQL → DuckDB)",
     description=(
         "Uses OpenAI’s GPT-3.5-Turbo to translate your question into SQL, "