Update app.py

Extract each customer message independently

app.py

@@ -61,6 +61,30 @@ Critical rules:
 - Always set reply_drafts to [].
 """
 
+SINGLE_ORDER_PROMPT = """
+You extract one order from one customer's DM.
+
+Return only valid JSON with this exact shape:
+{
+  "item": "",
+  "quantity": "",
+  "flavor": "",
+  "pickup_time": "",
+  "delivery_address": "",
+  "payment_status": "",
+  "notes": "",
+  "missing_fields": []
+}
+
+Rules:
+- Use only facts from this one message.
+- Do not invent details.
+- Put pickup or delivery time in pickup_time.
+- Put pickup place or delivery address in delivery_address.
+- If unknown, use an empty string.
+- missing_fields can only contain: quantity, flavor, pickup_time, delivery_address, payment_status.
+"""
+
 
 EXAMPLE_INPUT = """Maya: Hi! Can I get 2 dozen cupcakes for Saturday morning? Half vanilla, half chocolate.
 Sam: Need 1 birthday cake, chocolate, for pickup Friday 5pm. I can pay Venmo.
@@ -181,14 +205,40 @@ def build_reply_drafts(data):
 
     return replies
 
-def analyze_messages(messages):
-    if not messages.strip():
-        return pd.DataFrame(columns=ORDER_COLUMNS), "Paste some DMs first.", "", ""
+def split_customer_messages(messages):
+    entries = []
+    current_customer = ""
+    current_parts = []
+
+    for raw_line in messages.splitlines():
+        line = raw_line.strip()
+        if not line:
+            continue
+
+        if ":" in line:
+            possible_name, body = line.split(":", 1)
+            if possible_name.strip() and len(possible_name.strip().split()) <= 3:
+                if current_customer or current_parts:
+                    entries.append((current_customer or "Customer", " ".join(current_parts).strip()))
+                current_customer = possible_name.strip()
+                current_parts = [body.strip()]
+                continue
+
+        if current_parts:
+            current_parts.append(line)
+        else:
+            entries.append(("Customer", line))
+
+    if current_customer or current_parts:
+        entries.append((current_customer or "Customer", " ".join(current_parts).strip()))
 
+    return [(name, body) for name, body in entries if body]
+
+def extract_single_order(customer, message):
     prompt = tokenizer.apply_chat_template(
         [
-            {"role": "system", "content": SYSTEM_PROMPT},
-            {"role": "user", "content": messages},
+            {"role": "system", "content": SINGLE_ORDER_PROMPT},
+            {"role": "user", "content": f"Customer: {customer}\nMessage: {message}"},
         ],
         tokenize=False,
         add_generation_prompt=True,
@@ -198,7 +248,7 @@ def analyze_messages(messages):
     with torch.no_grad():
         output = model.generate(
             **inputs,
-            max_new_tokens=900,
+            max_new_tokens=350,
             do_sample=False,
             pad_token_id=tokenizer.eos_token_id,
         )
@@ -209,22 +259,50 @@ def analyze_messages(messages):
     )
 
     try:
-        data = extract_json(generated)
-    except Exception as exc:
-        return (
-            pd.DataFrame(columns=ORDER_COLUMNS),
-            f"### Needs review\nThe model did not return valid JSON: {exc}",
-            "",
-            generated,
-        )
+        parsed = extract_json(generated)
+    except Exception:
+        parsed = {
+            "item": "",
+            "quantity": "",
+            "flavor": "",
+            "pickup_time": "",
+            "delivery_address": "",
+            "payment_status": "",
+            "notes": message,
+            "missing_fields": [],
+        }
+
+    order = {"customer": customer}
+    for col in ORDER_COLUMNS[1:]:
+        value = parsed.get(col, "")
+        if col == "missing_fields":
+            if isinstance(value, list):
+                order[col] = value
+            elif isinstance(value, str):
+                order[col] = [part.strip() for part in value.split(",") if part.strip()]
+            else:
+                order[col] = []
+        else:
+            order[col] = text_value(value)
 
+    return order
+
+def analyze_messages(messages):
+    if not messages.strip():
+        return pd.DataFrame(columns=ORDER_COLUMNS), "Paste some DMs first.", "", ""
+
+    entries = split_customer_messages(messages)
+    orders_data = [extract_single_order(customer, message) for customer, message in entries]
+
+    data = {"orders": orders_data}
     orders_df = normalize_orders(data)
+
     auto_prep = build_prep_list(data)
     auto_replies = build_reply_drafts(data)
-    
+
     data["prep_list"] = auto_prep
     data["reply_drafts"] = auto_replies
-    
+
     prep = format_list("Prep list", auto_prep)
     replies = format_replies(auto_replies)
     raw = json.dumps(data, indent=2, ensure_ascii=False)