Update app.py

app.py

@@ -10,7 +10,6 @@ from langchain_community.chat_models import ChatOpenAI
 from langchain.agents import initialize_agent, Tool, AgentType
 from fuzzywuzzy import fuzz
 
-# --- CONFIGURATION ---
 st.set_page_config(page_title="Accounts Payable AI Agent", layout="wide")
 
 MODELS = {
@@ -21,27 +20,9 @@ MODELS = {
         "response_format": None,
         "extra_headers": {},
     },
-    # Keep other models for extraction/fine-tune if you want, but only OpenAI used for agent
-    "DeepSeek v3": {
-        "api_url": "https://api.deepseek.com/v1/chat/completions",
-        "model": "deepseek-chat",
-        "key_env": "DEEPSEEK_API_KEY",
-        "response_format": {"type": "json_object"},
-    },
-    "Mistral Small": {
-        "api_url": "https://openrouter.ai/api/v1/chat/completions",
-        "model": "mistralai/ministral-8b",
-        "key_env": "OPENROUTER_API_KEY",
-        "response_format": {"type": "json_object"},
-        "extra_headers": {
-            "HTTP-Referer": "https://huggingface.co",
-            "X-Title": "Invoice Extractor",
-        },
-    },
+    # You can add other models here for extraction/fine-tune if desired.
 }
 
-# --- UTILITY FUNCTIONS ---
-
 def get_api_key(model_choice):
     key = os.getenv(MODELS[model_choice]["key_env"])
     if not key:
@@ -231,16 +212,14 @@ def find_po_number_in_json(po_number, invoice_json):
             fields.append(str(obj))
         return fields
 
-    # Clean up PO number
     po_str = str(po_number).strip().replace(" ", "").replace(".0", "")
     try:
-        po_int = str(int(float(po_number)))  # also get int version
+        po_int = str(int(float(po_number)))
     except:
         po_int = po_str
 
     all_strs = [str(s).strip().replace(" ", "").replace(".0", "") for s in _flatten(invoice_json)]
 
-    # Check for exact match, or substring match, with all variants
     for s in all_strs:
         if not s:
             continue
@@ -250,13 +229,10 @@ def find_po_number_in_json(po_number, invoice_json):
             return True
     return False
 
-
 def find_best_po_match(inv, po_df):
     inv_hdr = inv["invoice_header"]
     inv_supplier = inv_hdr.get("supplier_name") or ""
     inv_po_number = inv_hdr.get("purchase_order_number") or inv_hdr.get("po_number") or inv_hdr.get("order_number") or ""
-    inv_ship_to = inv_hdr.get("ship_to_name") or ""
-    inv_bill_to = inv_hdr.get("bill_to_name") or ""
     inv_currency = inv_hdr.get("currency") or ""
     inv_total_due = clean_num(inv_hdr.get("total_due"))
     inv_line_items = inv.get("line_items", [])
@@ -264,9 +240,7 @@ def find_best_po_match(inv, po_df):
     scores = []
     for idx, row in po_df.iterrows():
         po_supplier = row.get("Supplier Name", "")
-        po_po_number = str(row.get("PO Number", ""))  # Make sure string
-        po_ship_to = row.get("Ship To", "")
-        po_bill_to = row.get("Bill To", "")
+        po_po_number = str(row.get("PO Number", ""))
         po_currency = row.get("Currency", "")
         po_total = clean_num(row.get("PO Total Value", ""))
         po_desc = row.get("Item Description", "")
@@ -284,7 +258,6 @@ def find_best_po_match(inv, po_df):
             "score": s_supplier
         })
 
-        # PO Number scoring: anywhere in JSON
         s_po_number = 100 if find_po_number_in_json(po_po_number, inv) else 0
         field_details.append({
             "field": "PO Number (anywhere in JSON)",
@@ -293,22 +266,6 @@ def find_best_po_match(inv, po_df):
             "score": s_po_number
         })
 
-        s_ship_to = weighted_fuzzy_score(inv_ship_to, po_ship_to)
-        field_details.append({
-            "field": "Ship To",
-            "invoice": inv_ship_to,
-            "po": po_ship_to,
-            "score": s_ship_to
-        })
-
-        s_bill_to = weighted_fuzzy_score(inv_bill_to, po_bill_to)
-        field_details.append({
-            "field": "Bill To",
-            "invoice": inv_bill_to,
-            "po": po_bill_to,
-            "score": s_bill_to
-        })
-
         s_currency = weighted_fuzzy_score(inv_currency, po_currency)
         field_details.append({
             "field": "Currency",
@@ -325,7 +282,7 @@ def find_best_po_match(inv, po_df):
             "score": s_total
         })
 
-        # Check for at least one line item strong match
+        # Line item logic as before
         line_item_score = 0
         line_reason = ""
         best_line_detail = None
@@ -363,22 +320,18 @@ def find_best_po_match(inv, po_df):
                     f"unit_score={unit_score}, amount_score={amount_score}"
                 )
 
-        # Adjust scoring weights: Supplier 20%, PO Number 15%, Ship To 10%, Bill To 10%, Currency 10%, Total Due 20%, Line Item 15%
+        # Adjust scoring weights: Supplier 25%, PO Number 25%, Currency 10%, Total Due 20%, Line Item 20%
         total_score = (
-            s_supplier * 0.20 +
-            s_po_number * 0.15 +
-            s_ship_to * 0.10 +
-            s_bill_to * 0.10 +
+            s_supplier * 0.25 +
+            s_po_number * 0.25 +
             s_currency * 0.10 +
             s_total * 0.20 +
-            line_item_score * 0.15
+            line_item_score * 0.20
         )
 
         reason = (
             f"Supplier match: {s_supplier}/100 (invoice: '{inv_supplier}' vs PO: '{po_supplier}'), "
             f"PO Number: {s_po_number}/100 ({'found anywhere in JSON' if s_po_number else 'not found'}), "
-            f"Ship To: {s_ship_to}/100 (invoice: '{inv_ship_to}' vs PO: '{po_ship_to}'), "
-            f"Bill To: {s_bill_to}/100 (invoice: '{inv_bill_to}' vs PO: '{po_bill_to}'), "
             f"Currency: {s_currency}/100 (invoice: '{inv_currency}' vs PO: '{po_currency}'), "
             f"Total Due: {'match' if s_total else 'no match'} (invoice: {inv_total_due} vs PO: {po_total}), "
             f"Line item best match: {int(line_item_score)}/100. {line_reason}"
@@ -388,8 +341,6 @@ def find_best_po_match(inv, po_df):
             "po_idx": idx,
             "po_supplier": po_supplier,
             "po_po_number": po_po_number,
-            "po_ship_to": po_ship_to,
-            "po_bill_to": po_bill_to,
             "po_total": po_total,
             "scores": field_details,
             "line_item_score": line_item_score,
@@ -407,7 +358,6 @@ def find_best_po_match(inv, po_df):
     best_row, best_score, reason, debug = scores[0]
     return best_row, best_score, reason, debug
 
-
 def extract_invoice_info(model_choice, text):
     prompt = get_extraction_prompt(model_choice, text)
     raw = query_llm(model_choice, prompt)
@@ -533,6 +483,11 @@ if st.button("Extract") and inv_file:
 extracted_info = st.session_state.get('last_extracted_info', None)
 po_df = st.session_state.get('last_po_df', None)
 
+# UI for scoring thresholds
+st.sidebar.header("Set Decision Thresholds")
+approved_threshold = st.sidebar.slider("Threshold for 'APPROVED'", min_value=0, max_value=100, value=85)
+partial_threshold = st.sidebar.slider("Threshold for 'PARTIALLY APPROVED'", min_value=0, max_value=approved_threshold-1, value=70)
+
 def po_match_tool_func(input_text):
     invoice = st.session_state.get("last_extracted_info")
     po_df = st.session_state.get("last_po_df")
@@ -545,9 +500,9 @@ def po_match_tool_func(input_text):
 
     best_row, best_score, reason, debug = find_best_po_match(invoice, po_df)
 
-    if best_score > 85:
+    if best_score > approved_threshold:
         status = "APPROVED"
-    elif best_score > 70:
+    elif best_score > partial_threshold:
         status = "PARTIALLY APPROVED"
     else:
         status = "REJECTED"
@@ -564,7 +519,7 @@ if po_df is not None:
 
 if extracted_info is not None and po_df is not None:
     st.markdown("---")
-    st.subheader("EZOFIS AP AGENT Decision")
+    st.subheader("EZOFIS AP AGENT Decision (OpenAI Only)")
     if st.button("Make a decision (EZOFIS AP AGENT)"):
         tools = [
             Tool(
@@ -573,7 +528,6 @@ if extracted_info is not None and po_df is not None:
                 description="Smartly match invoice to PO using all possible fields.",
             )
         ]
-        # Always use OpenAI GPT-4.1 for agent reasoning
         decision_llm = ChatOpenAI(
             openai_api_key=get_api_key("OpenAI GPT-4.1"),
             model=MODELS["OpenAI GPT-4.1"]["model"],
@@ -588,7 +542,7 @@ if extracted_info is not None and po_df is not None:
         )
         prompt = (
             "You are an expert accounts payable agent. "
-            "Use po_match_tool to check for the best possible match using supplier, PO number (which may appear anywhere in the invoice JSON, even within other fields), ship to, bill to, currency, line items, and total value. "
+            "Use po_match_tool to check for the best possible match using supplier, PO number (which may appear anywhere in the invoice JSON, even within other fields), currency, line items, and total value. "
             "Weigh the importance of each field as an expert would. "
             "Return a JSON with decision (APPROVED, PARTIALLY APPROVED, REJECTED), reason (include field scores and reasoning), debug, and the best matched PO row.\n"
             f"Invoice JSON:\n{json.dumps(extracted_info, indent=2)}"