Update app.py

app.py

@@ -3,16 +3,17 @@ import requests
 import os
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 
-model_id = "HuggingFaceH4/zephyr-7b-beta"  # Change here!
+# ---- Hugging Face model: use a public chat/instruct model ----
+model_id = "HuggingFaceH4/zephyr-7b-beta"  # You can use another public instruct/chat model if you prefer
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
+model = AutoModelForCausalLM.from_pretrained(model_id)
 llm = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-
 # --- SAP Sales Order Header tool ---
-def query_sales_order_header():
+def fetch_sales_order_headers():
+    """Fetches the top 5 SAP sales orders from the sandbox API and returns as a list of dicts."""
     api_url = "https://sandbox.api.sap.com/s4hanacloud/sap/opu/odata/sap/API_SALES_ORDER_SRV/A_SalesOrder?$top=5&$inlinecount=allpages"
-    api_key = os.getenv("SAP_SANDBOX_API_KEY", "YOUR_API_KEY")
+    api_key = os.getenv("SAP_SANDBOX_API_KEY", "YOUR_API_KEY")  # Set in Space secrets!
     headers = {
         "APIKey": api_key,
         "Accept": "application/json"
@@ -23,45 +24,60 @@ def query_sales_order_header():
         data = r.json()
         results = data.get('d', {}).get('results', [])
         if not results:
-            return "No sales orders found."
-        summaries = []
-        for order in results:
-            summaries.append(
-                f"Order: {order['SalesOrder']} | "
-                f"Type: {order['SalesOrderType']} | "
-                f"Org: {order['SalesOrganization']} | "
-                f"Date: {order['SalesOrderDate']} | "
-                f"SoldTo: {order['SoldToParty']} | "
-                f"Net: {order['TotalNetAmount']} {order['TransactionCurrency']} | "
-                f"Status: {order['OverallSDProcessStatus']}"
-            )
-        return "\n".join(summaries)
+            return []
+        return results
     except Exception as e:
         return f"Error fetching Sales Orders: {e}"
 
-# --- Main chat function ---
+def format_sales_orders_for_llm(orders):
+    """Converts the SAP sales order list to readable context for the LLM."""
+    if isinstance(orders, str):
+        return orders  # Pass through error message
+    context = "Here are the latest SAP sales orders:\n"
+    for i, order in enumerate(orders, 1):
+        context += (
+            f"{i}. Order: {order['SalesOrder']}, "
+            f"Type: {order['SalesOrderType']}, "
+            f"Org: {order['SalesOrganization']}, "
+            f"Date: {order['SalesOrderDate']}, "
+            f"SoldTo: {order['SoldToParty']}, "
+            f"Net: {order['TotalNetAmount']} {order['TransactionCurrency']}, "
+            f"Status: {order['OverallSDProcessStatus']}\n"
+        )
+    return context
+
 def chat_agent(message, history):
-    # Rudimentary tool use: check for sales order question
-    tool_trigger = ["sales order", "sap order", "sap sales"]
-    if any(kw in message.lower() for kw in tool_trigger):
-        tool_response = query_sales_order_header()
-        response = f"(SAP Sales Orders)\n{tool_response}"
-    else:
-        # Pure LLM response
-        response = llm(message, max_new_tokens=256)[0]["generated_text"]
+    # Step 1: Fetch SAP data every time
+    sales_orders = fetch_sales_order_headers()
+    context = format_sales_orders_for_llm(sales_orders)
+    # Step 2: Prompt LLM with both the question and the context
+    prompt = (
+        f"{context}\n"
+        f"User asked: {message}\n"
+        "Based on the above SAP sales orders, answer the user's question as accurately as possible. "
+        "If the question asks for sorting, filtering, maximum values, customer-specific info, or other details, do it based on the orders above. "
+        "If the question is unclear, summarize the available sales orders."
+    )
+    # Step 3: Generate LLM answer
+    llm_output = llm(prompt, max_new_tokens=256)[0]["generated_text"]
+    # Remove the prompt from the response if present
+    response = llm_output.replace(prompt, "").strip()
     history = history or []
     history.append((message, response))
     return history, history
 
-# --- Gradio UI ---
+# ---- Gradio UI ----
 with gr.Blocks() as demo:
     gr.Markdown(
         """
-        # SAP Sales Order Chat Agent  
-        Ask about SAP sales orders or general questions.  
-        Example:  
-        - `Show me the latest SAP sales orders.`  
-        - `Tell me about SAP order values.`  
+        # SAP Sales Order Chat Agent (Smart Reasoning)
+        Ask about SAP sales orders, values, filtering, sorting, etc.  
+        Example questions:
+        - `Show me the latest SAP sales orders.`
+        - `Tell me about SAP order values.`
+        - `Show me top 2 sales orders with maximum value.`
+        - `Which order has the highest net value?`
+        - `List orders by customer.`
         """
     )
     chatbot = gr.Chatbot()