Update app.py

app.py

@@ -10,70 +10,50 @@ from langchain_community.chat_models import ChatOpenAI
 from langchain.agents import initialize_agent, Tool, AgentType
 from fuzzywuzzy import fuzz
 
-# ---- Custom CSS for SaaS look ----
+# --- Streamlit Page Settings ---
+st.set_page_config(page_title="EZOFIS Accounts Payable Agent", layout="wide")
+
+# --- Styles for SaaS Feel ---
 st.markdown("""
     <style>
-        #MainMenu, header, footer {visibility: hidden;}
-        html, body, [class*="css"]  {
-            font-family: 'Inter', 'Segoe UI', Arial, sans-serif !important;
-            background: #f7f8fa !important;
-        }
-        .app-title {
-            font-size: 2.2rem;
-            font-weight: 800;
-            margin-bottom: 0.5rem;
-            letter-spacing: -1px;
-        }
-        .block-card {
-            background: #fff;
-            border-radius: 16px;
-            box-shadow: 0 4px 24px rgba(30,34,90,0.04);
-            padding: 2.2rem 2rem 2rem 2rem;
-            margin-bottom: 2rem;
-        }
-        .step-num {
-            color: #fff;
-            background: #5356FB;
-            font-weight: 700;
-            font-size: 1rem;
-            display: inline-block;
-            border-radius: 100px;
-            padding: 0.45em 0.85em;
-            margin-right: 0.75em;
-        }
-        .stButton button {
-            border-radius: 100px !important;
-            background: #5356FB !important;
-            color: #fff !important;
-            font-weight: 600;
-            border: none;
-            box-shadow: 0 2px 8px rgba(30,34,90,0.05);
-            padding: 0.6em 2em !important;
-            transition: all 0.18s;
-        }
-        .stButton button:hover {
-            background: #2123A6 !important;
-            color: #fff !important;
-            transform: translateY(-1px) scale(1.03);
-        }
-        .stSlider > div {padding-top: 0.7em;}
-        .stSlider label {font-size: 1.07em;}
-        .stFileUploader > div { border-radius: 30px !important; }
-        .stFileUploader label {font-size: 1.09em;}
-        .stStatusWidget {display: none !important;}
-        /* Remove success box border (for record count) */
-        .element-container .stAlert {box-shadow:none !important;border:none !important;}
+    .block-card {
+        background: #fff;
+        border-radius: 20px;
+        box-shadow: 0 2px 16px rgba(25,39,64,0.05);
+        padding: 32px 26px 24px 26px;
+        margin-bottom: 24px;
+    }
+    .step-num {
+        background: #2F49D1;
+        color: #fff;
+        border-radius: 999px;
+        padding: 6px 13px;
+        font-weight: 700;
+        margin-right: 14px;
+        font-size: 20px;
+        display: inline-block;
+        vertical-align: middle;
+    }
+    .stButton>button {
+        background: #2F49D1 !important;
+        color: white !important;
+        border-radius: 12px !important;
+        padding: 10px 32px !important;
+        font-weight: 700;
+        border: none !important;
+        font-size: 18px !important;
+        margin-top: 12px !important;
+    }
+    .stSlider>div>div>div>div {
+        background: #F3F6FB !important;
+        border-radius: 999px;
+    }
+    .css-12w0qpk {padding-top: 0rem;} /* Removes extra padding */
+    .css-1kyxreq {padding-top: 0rem;} /* Removes extra padding */
     </style>
 """, unsafe_allow_html=True)
 
-st.markdown('<div class="app-title">EZOFIS Accounts Payable Agent</div>', unsafe_allow_html=True)
-st.markdown('<span style="color:#444; font-size:1.13em;">Modern workflow automation for finance teams</span>', unsafe_allow_html=True)
-st.write("")
-
-# -- Layout: cards in two columns --
-col1, col2 = st.columns([1, 1.15], gap="large")
-
-# -- Business logic variables
+# --- Model Definitions ---
 MODELS = {
     "OpenAI GPT-4.1": {
         "api_url": "https://api.openai.com/v1/chat/completions",
@@ -83,7 +63,6 @@ MODELS = {
         "extra_headers": {},
     },
 }
-mdl = "OpenAI GPT-4.1"
 
 def get_api_key(model_choice):
     key = os.getenv(MODELS[model_choice]["key_env"])
@@ -162,7 +141,7 @@ def get_extraction_prompt(model_choice, txt):
         "Use this schema:\n"
         '{\n'
         '  "invoice_header": {...},\n'
-        '  "line_items": [ ... ]\n'
+        '  "line_items": [{...}]\n'
         '}'
         "\nIf a field is missing for a line item or header, use null. "
         "Do not invent fields. Do not add any header or shipment data to any line item. Return ONLY the JSON object, no explanation.\n"
@@ -171,8 +150,9 @@ def get_extraction_prompt(model_choice, txt):
     )
 
 def ensure_total_due(invoice_header):
+    # Prefer total_before_tax if total_due mismatches in scoring
     if invoice_header.get("total_due") in [None, ""]:
-        for field in ["invoice_total", "invoice_value", "total_before_tax", "balance_due", "amount_paid"]:
+        for field in ["total_before_tax", "invoice_total", "invoice_value", "balance_due", "amount_paid"]:
             if field in invoice_header and invoice_header[field]:
                 invoice_header["total_due"] = invoice_header[field]
                 break
@@ -208,11 +188,13 @@ def find_po_number_in_json(po_number, invoice_json):
         elif obj is not None:
             fields.append(str(obj))
         return fields
+
     po_str = str(po_number).strip().replace(" ", "").replace(".0", "")
     try:
         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)]
     for s in all_strs:
         if not s:
@@ -228,8 +210,11 @@ def find_best_po_match(inv, po_df, weight_supplier, weight_po_number, weight_cur
     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_currency = inv_hdr.get("currency") or ""
+    # -- Try both total_due and total_before_tax for matching
     inv_total_due = clean_num(inv_hdr.get("total_due"))
+    inv_total_before_tax = clean_num(inv_hdr.get("total_before_tax"))
     inv_line_items = inv.get("line_items", [])
+
     scores = []
     for idx, row in po_df.iterrows():
         po_supplier = row.get("Supplier Name", "")
@@ -240,15 +225,47 @@ def find_best_po_match(inv, po_df, weight_supplier, weight_po_number, weight_cur
         po_qty = str(row.get("Item Quantity", ""))
         po_unit = str(row.get("Item Unit Price", ""))
         po_line_total = clean_num(row.get("Line Item Total", ""))
+
         field_details = []
+
         s_supplier = weighted_fuzzy_score(inv_supplier, po_supplier)
-        field_details.append({"field": "Supplier Name","invoice": inv_supplier,"po": po_supplier,"score": s_supplier})
+        field_details.append({
+            "field": "Supplier Name",
+            "invoice": inv_supplier,
+            "po": po_supplier,
+            "score": s_supplier
+        })
+
         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)","invoice": "found" if s_po_number else "not found","po": po_po_number,"score": s_po_number})
+        field_details.append({
+            "field": "PO Number (anywhere in JSON)",
+            "invoice": "found" if s_po_number else "not found",
+            "po": po_po_number,
+            "score": s_po_number
+        })
+
         s_currency = weighted_fuzzy_score(inv_currency, po_currency)
-        field_details.append({"field": "Currency","invoice": inv_currency,"po": po_currency,"score": s_currency})
-        s_total = 100 if inv_total_due is not None and po_total is not None and abs(inv_total_due - po_total) < 2 else 0
-        field_details.append({"field": "Total Due","invoice": inv_total_due,"po": po_total,"score": s_total})
+        field_details.append({
+            "field": "Currency",
+            "invoice": inv_currency,
+            "po": po_currency,
+            "score": s_currency
+        })
+
+        # Try total_due, fallback to total_before_tax
+        s_total = 0
+        if inv_total_due is not None and po_total is not None and abs(inv_total_due - po_total) < 2:
+            s_total = 100
+        elif inv_total_before_tax is not None and po_total is not None and abs(inv_total_before_tax - po_total) < 2:
+            s_total = 100
+        field_details.append({
+            "field": "Total Due or Before Tax",
+            "invoice": inv_total_due if s_total else inv_total_before_tax,
+            "po": po_total,
+            "score": s_total
+        })
+
+        # Line item logic as before
         line_item_score = 0
         line_reason = ""
         best_line_detail = None
@@ -281,7 +298,11 @@ def find_best_po_match(inv, po_df, weight_supplier, weight_po_number, weight_cur
             if total > line_item_score:
                 line_item_score = total
                 best_line_detail = detail
-                line_reason = (f"Best line item: desc_score={desc_score}, qty_score={qty_score}, unit_score={unit_score}, amount_score={amount_score}")
+                line_reason = (
+                    f"Best line item: desc_score={desc_score}, qty_score={qty_score}, "
+                    f"unit_score={unit_score}, amount_score={amount_score}"
+                )
+
         wsum = weight_supplier + weight_po_number + weight_currency + weight_total_due + weight_line_item
         total_score = (
             s_supplier * weight_supplier/100 +
@@ -290,13 +311,15 @@ def find_best_po_match(inv, po_df, weight_supplier, weight_po_number, weight_cur
             s_total * weight_total_due/100 +
             line_item_score * weight_line_item/100
         ) if wsum == 100 else 0
+
         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"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"Total: {'match' if s_total else 'no match'} (invoice: {inv_total_due if s_total else inv_total_before_tax} vs PO: {po_total}), "
             f"Line item best match: {int(line_item_score)}/100. {line_reason}"
         )
+
         debug = {
             "po_idx": idx,
             "po_supplier": po_supplier,
@@ -307,9 +330,11 @@ def find_best_po_match(inv, po_df, weight_supplier, weight_po_number, weight_cur
             "best_line_detail": best_line_detail,
             "total_score": total_score,
             "line_reason": line_reason,
-            "inv_total_due": inv_total_due
+            "inv_total_due": inv_total_due,
+            "inv_total_before_tax": inv_total_before_tax
         }
         scores.append((row, total_score, reason, debug))
+
     scores.sort(key=lambda tup: tup[1], reverse=True)
     if not scores:
         return None, 0, "No POs found.", {}
@@ -401,14 +426,24 @@ def extract_text_from_unstract(uploaded_file):
     except Exception:
         return r.text
 
-# ---- UI/Workflow in columns ----
+# ---------------- UI LAYOUT ----------------------
+st.markdown(
+    "<h1 style='font-weight:800; margin-bottom:8px;'>EZOFIS Accounts Payable Agent</h1>",
+    unsafe_allow_html=True
+)
+st.markdown(
+    "<div style='font-size:20px; margin-bottom:28px; color:#24345C;'>Modern workflow automation for finance teams</div>",
+    unsafe_allow_html=True
+)
+
+# ---- Three columns layout for horizontal flow
+col1, col2, col3 = st.columns([2,2,3])
 
+# ---- Step 1: Upload POs (col1) ----
 with col1:
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">1</span> <b>Upload Active Purchase Orders (POs)</b>', unsafe_allow_html=True)
-    st.markdown('<span style="color:#888;">CSV with PO number, Supplier, Items, etc.</span>', unsafe_allow_html=True)
+    st.markdown("<span class='step-num'>1</span> <b>Upload Active Purchase Orders (POs)</b>", unsafe_allow_html=True)
     po_file = st.file_uploader(
-        "Drop your POs CSV here (max 200MB)",
+        "CSV with PO number, Supplier, Items, etc.",
         type=["csv"],
         key="po_csv",
         label_visibility="collapsed"
@@ -418,11 +453,11 @@ with col1:
         po_df = pd.read_csv(po_file)
         st.success(f"Loaded {len(po_df)} records from uploaded CSV.")
         st.session_state['last_po_df'] = po_df
-    st.markdown("</div>", unsafe_allow_html=True)
 
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">2</span> <b>Configure Scoring Weights</b>', unsafe_allow_html=True)
-    st.markdown('<span style="color:#888;">Set weights for matching. Total must equal 100%.</span>', unsafe_allow_html=True)
+# ---- Step 2: Scoring Weights (col1) ----
+with col1:
+    st.markdown("<span class='step-num'>2</span> <b>Configure Scoring Weights</b>", unsafe_allow_html=True)
+    st.markdown("Set weights for matching. Total must equal 100%.", unsafe_allow_html=True)
     def int_slider(label, value, key):
         return st.slider(label, 0, 100, value, 1, key=key, format="%d")
     weight_supplier = int_slider("Supplier Name (%)", 25, "w_supplier")
@@ -433,74 +468,70 @@ with col1:
     weight_sum = weight_supplier + weight_po_number + weight_currency + weight_total_due + weight_line_item
     if weight_sum != 100:
         st.warning(f"Sum of weights is {weight_sum}%. Adjust so it equals 100%.")
-    st.markdown("</div>", unsafe_allow_html=True)
 
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">3</span> <b>Decision Thresholds</b>', unsafe_allow_html=True)
+    st.markdown("<span class='step-num'>3</span> <b>Set Decision Thresholds</b>", unsafe_allow_html=True)
     approved_threshold = st.slider("Threshold for 'APPROVED'", min_value=0, max_value=100, value=85, format="%d")
     partial_threshold = st.slider("Threshold for 'PARTIALLY APPROVED'", min_value=0, max_value=approved_threshold-1, value=70, format="%d")
-    st.markdown("</div>", unsafe_allow_html=True)
 
+# ---- Step 4: Upload Invoice (col2) ----
 with col2:
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">4</span> <b>Upload Invoice/Document</b>', unsafe_allow_html=True)
+    st.markdown("<span class='step-num'>4</span> <b>Upload Invoice/Document</b>", unsafe_allow_html=True)
     inv_file = st.file_uploader(
-        "Drop invoice or document here (PDF, Excel, Image, Word, etc.)",
+        "Upload PDF, DOCX, XLSX, PNG, JPG, TIFF",
         type=["pdf", "docx", "xlsx", "xls", "png", "jpg", "jpeg", "tiff"],
-        key="inv_file",
+        key="invoice_file",
         label_visibility="collapsed"
     )
-    st.markdown("</div>", unsafe_allow_html=True)
-
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">5</span> <b>Extract Data</b>', unsafe_allow_html=True)
-    if st.button("Extract") and inv_file:
-        with st.spinner("Extracting text from document..."):
-            text = extract_text_from_unstract(inv_file)
-        if text:
-            extracted_info = extract_invoice_info(mdl, text)
-            if extracted_info:
-                if "invoice_header" in extracted_info:
-                    extracted_info["invoice_header"] = ensure_total_due(extracted_info["invoice_header"])
-                st.success("Extraction Complete")
-                st.markdown("**Invoice Metadata**")
-                st.json(extracted_info["invoice_header"], expanded=False)
-                st.markdown("**Line Items**")
-                st.json(extracted_info["line_items"], expanded=False)
-                st.session_state['last_extracted_info'] = extracted_info
-    st.markdown("</div>", unsafe_allow_html=True)
-
-    # AP Decision Card
-    st.markdown('<div class="block-card">', unsafe_allow_html=True)
-    st.markdown('<span class="step-num">6</span> <b>AP Agent Decision</b>', unsafe_allow_html=True)
-    extracted_info = st.session_state.get('last_extracted_info', None)
-    po_df = st.session_state.get('last_po_df', None)
-    def po_match_tool_func(input_text):
-        invoice = st.session_state.get("last_extracted_info")
-        po_df = st.session_state.get("last_po_df")
-        if invoice is None or po_df is None:
-            return json.dumps({
-                "decision": "REJECTED",
-                "reason": "Invoice or PO data not found.",
-                "debug": {},
-            })
-        best_row, best_score, reason, debug = find_best_po_match(
-            invoice, po_df, weight_supplier, weight_po_number, weight_currency, weight_total_due, weight_line_item
-        )
-        if best_score > approved_threshold:
-            status = "APPROVED"
-        elif best_score > partial_threshold:
-            status = "PARTIALLY APPROVED"
+
+# ---- Step 5: Extract Data (col2) ----
+with col2:
+    st.markdown("<span class='step-num'>5</span> <b>Extract Data</b>", unsafe_allow_html=True)
+    if st.button("Extract"):
+        if inv_file:
+            with st.spinner("Extracting text from document..."):
+                text = extract_text_from_unstract(inv_file)
+            if text:
+                mdl = "OpenAI GPT-4.1"
+                extracted_info = extract_invoice_info(mdl, text)
+                if extracted_info:
+                    if "invoice_header" in extracted_info:
+                        extracted_info["invoice_header"] = ensure_total_due(extracted_info["invoice_header"])
+                    st.success("Extraction Complete")
+                    st.session_state['last_extracted_info'] = extracted_info
         else:
-            status = "REJECTED"
-        return json.dumps({
-            "decision": status,
-            "reason": f"Best match score: {int(best_score)}/100. {reason}",
-            "debug": debug,
-            "po_row": best_row.to_dict() if best_row is not None else None
-        })
-    if extracted_info is not None and po_df is not None:
-        if st.button("Make a decision (EZOFIS AP AGENT)"):
+            st.warning("Please upload an invoice/document first.")
+
+# ---- Step 6: AP Agent Decision (col3) ----
+with col3:
+    st.markdown("<span class='step-num'>6</span> <b>AP Agent Decision</b>", unsafe_allow_html=True)
+    if st.button("Make a decision (EZOFIS AP AGENT)"):
+        extracted_info = st.session_state.get('last_extracted_info', None)
+        po_df = st.session_state.get('last_po_df', None)
+        if extracted_info is not None and po_df is not None:
+            def po_match_tool_func(input_text):
+                invoice = st.session_state.get("last_extracted_info")
+                po_df = st.session_state.get("last_po_df")
+                if invoice is None or po_df is None:
+                    return json.dumps({
+                        "decision": "REJECTED",
+                        "reason": "Invoice or PO data not found.",
+                        "debug": {},
+                    })
+                best_row, best_score, reason, debug = find_best_po_match(
+                    invoice, po_df, weight_supplier, weight_po_number, weight_currency, weight_total_due, weight_line_item
+                )
+                if best_score > approved_threshold:
+                    status = "APPROVED"
+                elif best_score > partial_threshold:
+                    status = "PARTIALLY APPROVED"
+                else:
+                    status = "REJECTED"
+                return json.dumps({
+                    "decision": status,
+                    "reason": f"Best match score: {int(best_score)}/100. {reason}",
+                    "debug": debug,
+                    "po_row": best_row.to_dict() if best_row is not None else None
+                })
             tools = [
                 Tool(
                     name="po_match_tool",
@@ -509,8 +540,8 @@ with col2:
                 )
             ]
             decision_llm = ChatOpenAI(
-                openai_api_key=get_api_key(mdl),
-                model=MODELS[mdl]["model"],
+                openai_api_key=get_api_key("OpenAI GPT-4.1"),
+                model=MODELS["OpenAI GPT-4.1"]["model"],
                 temperature=0,
                 streaming=False,
             )
@@ -529,17 +560,24 @@ with col2:
             )
             with st.spinner("AI is reasoning and making a decision..."):
                 result = agent.run(prompt)
+            # Always display debug/info
+            st.markdown("<h3 style='margin-top:18px;'>AI Decision & Reason</h3>", unsafe_allow_html=True)
             try:
                 result_json = json.loads(result)
                 st.write(f"**Decision:** {result_json.get('decision', 'N/A')}")
                 st.write(f"**Reason:** {result_json.get('reason', 'N/A')}")
-                with st.expander("Debug & Matching Details"):
-                    st.json(result_json.get('debug'))
-                    st.subheader("Extracted Invoice JSON")
-                    st.json(extracted_info)
-                    st.subheader("Matched PO Row")
-                    st.json(result_json.get('po_row'))
+                st.markdown("##### Debug & Matching Details")
+                st.json(result_json.get('debug'))
+                st.markdown("##### Extracted Invoice JSON")
+                st.json(extracted_info)
+                st.markdown("##### Matched PO Row")
+                st.json(result_json.get('po_row'))
             except Exception:
                 st.subheader("AI Decision & Reason")
                 st.write(result)
-    st.markdown("</div>", unsafe_allow_html=True)
+
+# Always show extraction/decision debug in full for troubleshooting
+if "last_api" in st.session_state:
+    with st.expander("Debug"):
+        st.code(st.session_state.last_api)
+        st.code(st.session_state.last_raw)