feat: implement deterministic explanation and conditional Gemini advice, refactor risk scoring, and update UI elements and labels

app.py

@@ -6,19 +6,21 @@ import plotly.graph_objects as go
 from google import genai
 from transformers import pipeline
 
-# --- KONFIGURASI AI ---
+# --- KONFIGURASI GENERATIVE AI ---
 GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
-client = genai.Client(api_key=GOOGLE_API_KEY)
+# Inisialisasi Client (SDK Terbaru 2026)
+client_ai = genai.Client(api_key=GOOGLE_API_KEY) if GOOGLE_API_KEY else None
 
-# --- PALETTE WARNA & CSS CUSTOM ---
-# Primary: #0514DE, Secondary: #82C3EB, Pale: #E0EDF4, Accent: #F7BD87, White: #FFFFFF
+# --- PALET WARNA BANK NAGARI ---
+# #0514DE (Deep Blue), #82C3EB (Light Blue), #E0EDF4 (Pale), #F7BD87 (Gold), #FFFFFF (White)
 custom_css = """
 @import url('https://fonts.googleapis.com/css2?family=Plus+Jakarta+Sans:wght@400;600;700&display=swap');
 .gradio-container { font-family: 'Plus Jakarta Sans', sans-serif !important; background-color: #FFFFFF !important; }
 .nagari-header { background: linear-gradient(135deg, #0514DE 0%, #82C3EB 100%); color: white; padding: 30px; border-radius: 15px; border-bottom: 6px solid #F7BD87; margin-bottom: 25px; text-align: center; }
-.card-style { background: #E0EDF4; border-radius: 12px; padding: 20px; border: 1px solid #82C3EB; box-shadow: 2px 4px 10px rgba(5, 20, 222, 0.05); }
-.status-badge { display: inline-block; padding: 5px 15px; border-radius: 20px; font-weight: bold; color: white; }
-.advice-box { background: white; border-left: 5px solid #F7BD87; padding: 20px; border-radius: 10px; font-size: 1.05em; line-height: 1.6; }
+.card-nagari { background: #E0EDF4; border-radius: 12px; padding: 20px; border: 1px solid #82C3EB; box-shadow: 0 4px 10px rgba(5, 20, 122, 0.05); }
+.status-badge { display: inline-block; padding: 5px 15px; border-radius: 20px; font-weight: bold; color: white; margin-top: 10px; }
+.report-box { background: white; border-radius: 10px; padding: 20px; border-left: 6px solid #0514DE; line-height: 1.6; }
+.advice-box { background: #fffdf0; border: 1px solid #F7BD87; padding: 20px; border-radius: 10px; }
 """
 
 class ArchonNagariEngine:
@@ -28,115 +30,135 @@ class ArchonNagariEngine:
         except: self.classifier = None
 
     def load_data(self):
-        # Fix C0014: fillna memastikan data siap olah
+        # Fase 1: Foundation (Fix C0014 error handling)
         self.df_txn = pd.read_csv('transactions.csv', parse_dates=['date']).fillna({"raw_description": "Transaksi Umum"})
         self.df_cust = pd.read_csv('customers.csv')
         self.df_bal = pd.read_csv('balances_revised.csv', parse_dates=['month']).fillna(0)
         self.df_rep = pd.read_csv('repayments_revised.csv', parse_dates=['due_date']).fillna("on_time")
 
     def analyze(self, customer_id):
-        # Filter Data
+        # 1. Filter Data & Validation
         u_txn = self.df_txn[self.df_txn['customer_id'] == customer_id].copy()
         u_bal = self.df_bal[self.df_bal['customer_id'] == customer_id].sort_values('month')
         u_rep = self.df_rep[self.df_rep['customer_id'] == customer_id]
-        u_info = self.df_cust[self.df_cust['customer_id'] == customer_id].iloc[0]
         
         if u_txn.empty or u_bal.empty: return None
 
-        # --- FASE 4: RISK SCORING (WEIGHTED) ---
+        # --- FASE 4: RISK SCORING (DETERMINISTIC AI) ---
         income_txn = u_txn[u_txn['transaction_type'] == 'credit']['amount'].sum()
-        base_income = max(income_txn, u_info['monthly_income'])
+        cust_profile = self.df_cust[self.df_cust['customer_id'] == customer_id].iloc[0]
+        base_inc = max(income_txn, cust_profile['monthly_income'])
         expense = u_txn[u_txn['transaction_type'] == 'debit']['amount'].sum()
-        er = expense / base_income
+        er = expense / base_inc
         
         er_s = 1.0 if er > 0.8 else (0.5 if er > 0.5 else 0.0)
         bt_s = 1.0 if len(u_bal) >= 2 and u_bal.iloc[-1]['avg_balance'] < u_bal.iloc[-2]['avg_balance'] else 0.0
         od_s = 1.0 if (u_bal['min_balance'] <= 0).any() else 0.0
         mp_s = 1.0 if (u_rep['status'] == 'late').any() else 0.0
         
-        score = (0.3 * er_s) + (0.2 * bt_s) + (0.2 * od_s) + (0.2 * mp_s) + 0.1
-        risk_lv = "HIGH" if score >= 0.7 else ("MEDIUM" if score >= 0.4 else "LOW")
+        # Risk Score 0.0 - 1.0 (Bobot 30/20/20/20/10)
+        final_score = (0.3 * er_s) + (0.2 * bt_s) + (0.2 * od_s) + (0.2 * mp_s) + 0.1
+        risk_lv = "HIGH" if final_score >= 0.7 else ("MEDIUM" if final_score >= 0.4 else "LOW")
         
-        return risk_lv, score, er, u_bal, u_txn, base_income, expense
+        return risk_lv, final_score, er, u_bal, u_txn, base_inc, expense
 
-    def get_adaptive_advice(self, risk_lv, er, score, u_txn, base_income, expense):
-        # FASE 5 & 6: ADAPTIVE GEN-AI EXPLANATION
-        recent_txn = u_txn.tail(3)['raw_description'].tolist()
-        # Prompt yang lebih teknis dan mendalam agar Gemini tidak template
-        prompt = f"""
-        Identitas: Anda adalah Senior Financial Advisor Archon di Bank Nagari.
-        Analisis Data Nasabah:
-        - Status Risiko: {risk_lv} (Skor: {score:.2f}/1.00)
-        - Rasio Pengeluaran: {er:.2%} (Pengeluaran: Rp{expense:,.0f} dari Pendapatan: Rp{base_income:,.0f})
-        - Riwayat Belanja Terakhir: {', '.join(recent_txn)}
+    def get_deterministic_explanation(self, risk_lv, score, er, u_bal):
+        # Penjelasan adaptif berdasarkan output riil (Backup jika LLM gagal)
+        exp = f"### 📖 Interpretasi Analisis Archon\n"
+        if risk_lv == "HIGH":
+            exp += f"⚠️ **Level Risiko Tinggi ({score:.2f})**: Perilaku keuangan Anda memerlukan restrukturisasi segera. "
+        elif risk_lv == "MEDIUM":
+            exp += f"🔔 **Level Risiko Menengah ({score:.2f})**: Anda berada dalam zona peringatan dini (Early Warning). "
+        else:
+            exp += f"✅ **Level Risiko Rendah ({score:.2f})**: Resiliensi finansial Anda terpantau sangat sehat. "
+            
+        exp += f"\n\n**Rincian Metrik:**\n"
+        exp += f"- **Expense Ratio ({er:.1%})**: Anda menggunakan {er:.1%} pendapatan untuk belanja. "
+        exp += f"{'Sangat boros (>80%), kurangi pengeluaran lifestyle.' if er > 0.8 else 'Cukup aman, namun tetap waspada.'}\n"
+        
+        latest_bal = u_bal.iloc[-1]['avg_balance']
+        exp += f"- **Tren Saldo**: Saldo rata-rata Anda Rp{latest_bal:,.0f}. "
+        exp += f"{'Tren menurun dibanding bulan lalu.' if len(u_bal) > 1 and latest_bal < u_bal.iloc[-2]['avg_balance'] else 'Saldo terpantau stabil/naik.'}"
         
-        Tugas:
-        1. Jelaskan arti skor {score:.2f} dan level {risk_lv} secara singkat dan logis.
-        2. Berikan saran personal yang dihubungkan dengan pengeluaran {er:.2%} dan riwayat belanja tersebut.
-        3. Gunakan sapaan Bapak/Ibu, bahasa Indonesia yang hangat namun profesional.
-        4. Jangan gunakan pembukaan template, langsung ke inti analisis.
+        return exp
+
+    def get_gemini_advice(self, risk_lv, er, cust_id, u_txn):
+        # FASE 5: NBO GENERATIVE
+        if not client_ai: return None
+        merchants = u_txn.tail(3)['raw_description'].tolist()
+        prompt = f"""
+        Identitas: Senior Wealth Manager Bank Nagari.
+        Analisis Nasabah {cust_id}: Risiko {risk_lv}, Rasio Pengeluaran {er:.1%}.
+        Riwayat Belanja: {', '.join(merchants)}.
+        Tugas: Berikan saran finansial spesifik & empati untuk Bapak/Ibu. Hubungkan dengan pengeluaran di {merchants[0]}. Maks 3 kalimat.
         """
         try:
-            resp = client.models.generate_content(model="gemini-1.5-flash", contents=prompt)
+            resp = client_ai.models.generate_content(model="gemini-1.5-flash", contents=prompt)
             return resp.text
-        except:
-            return "Mohon maaf, sistem advisor sedang melakukan pemeliharaan. Silakan merujuk pada metrik risiko di panel kiri."
+        except: return None
 
-    def create_plots(self, u_bal, u_txn):
-        # Plot 1: Cashflow (Nagari Colors)
+    def create_viz(self, u_bal, u_txn):
+        # Plot 1: Cashflow (Fase 6)
         u_txn['m'] = u_txn['date'].dt.to_period('M').dt.to_timestamp()
         cf = u_txn.groupby(['m', 'transaction_type'])['amount'].sum().unstack().fillna(0)
         fig1 = go.Figure()
         fig1.add_trace(go.Bar(x=cf.index, y=cf.get('credit', 0), name='Pemasukan', marker_color='#82C3EB'))
         fig1.add_trace(go.Bar(x=cf.index, y=cf.get('debit', 0), name='Pengeluaran', marker_color='#0514DE'))
-        fig1.update_layout(title="Arus Kas Bulanan", barmode='group', template='plotly_white')
+        fig1.update_layout(title="Inflow vs Outflow Bulanan", barmode='group', template='plotly_white')
 
-        # Plot 2: Balance History
+        # Plot 2: Balance Trend
         fig2 = go.Figure()
         fig2.add_trace(go.Scatter(x=u_bal['month'], y=u_bal['avg_balance'], name='Saldo Rata-rata', line=dict(color='#F7BD87', width=4)))
-        fig2.update_layout(title="Tren Saldo", template='plotly_white')
+        fig2.update_layout(title="Kesehatan Saldo (Fase 6)", template='plotly_white')
         return fig1, fig2
 
 # --- UI LOGIC ---
 engine = ArchonNagariEngine()
 
-def process(cust_id):
+def run_archon(cust_id):
     res = engine.analyze(cust_id)
     if not res: return "## ❌ ID Tidak Valid", "Data tidak ditemukan.", None, None
     
     risk_lv, score, er, u_bal, u_txn, inc, exp = res
-    advice = engine.get_adaptive_advice(risk_lv, er, score, u_txn, inc, exp)
+    det_exp = engine.get_deterministic_explanation(risk_lv, score, er, u_bal)
+    gen_advice = engine.get_gemini_advice(risk_lv, er, cust_id, u_txn)
     f1, f2 = engine.create_plots(u_bal, u_txn)
     
+    # Gabungkan penjelasan Deterministik dengan Gemini (Jika Gemini ok)
+    final_report = det_exp
+    if gen_advice:
+        final_report += f"\n\n---\n**💡 Rekomendasi Virtual Advisor:**\n{gen_advice}"
+    else:
+        final_report += f"\n\n---\n*Saran Virtual Advisor sedang dalam sinkronisasi sistem.*"
+        
     color = "#ef4444" if risk_lv == "HIGH" else ("#f59e0b" if risk_lv == "MEDIUM" else "#10b981")
-    
     status_html = f"""
-    <div class='card-style'>
+    <div class='card-nagari'>
         <h2 style='color: #0514DE; margin:0;'>Hasil Analisis AI</h2>
-        <p style='margin:10px 0;'>Level Risiko: <span class='status-badge' style='background:{color}'>{risk_lv}</span></p>
-        <p><b>Risk Score:</b> {score:.2f} / 1.00</p>
+        <span class='status-badge' style='background:{color}'>{risk_lv} RISK</span>
+        <p style='margin-top:15px;'><b>Risk Score:</b> {score:.2f} / 1.00</p>
         <p><b>Expense Ratio:</b> {er:.1%}</p>
     </div>
     """
-    return status_html, advice, f1, f2
+    return status_html, final_report, f1, f2
 
 with gr.Blocks(css=custom_css) as demo:
-    gr.HTML("<div class='nagari-header'><h1>🛡️ ARCHON-AI: FINANCIAL ADVISOR</h1><p>Inteligensi Manajemen Risiko & Resiliensi Perbankan</p></div>")
+    gr.HTML("<div class='nagari-header'><h1>🛡️ ARCHON-AI: FINANCIAL ADVISOR</h1><p>Sistem Intelijen Resiliensi Finansial & Manajemen Risiko</p></div>")
     
     with gr.Row():
         with gr.Column(scale=1):
-            id_in = gr.Textbox(label="Customer ID", placeholder="C0001 - C0120")
-            btn = gr.Button("RUN ANALYSIS", variant="primary")
+            id_in = gr.Textbox(label="Customer ID", placeholder="Masukkan ID (C0001 - C0120)")
+            btn = gr.Button("PROSES DATA", variant="primary")
             out_status = gr.HTML()
             gr.Markdown("---")
-            gr.Markdown("**Interpretasi Metrik:**\n* **Risk Score**: Evaluasi gabungan saldo, cicilan, dan belanja.\n* **Expense Ratio**: Persentase gaji yang terpakai bulan ini.")
+            gr.Markdown("ℹ️ **Info Metrik**: Skor risiko didasarkan pada perpaduan saldo, rasio belanja, dan kedisiplinan cicilan nasabah.")
         
         with gr.Column(scale=2):
             with gr.Tabs():
-                with gr.TabItem("Inflow vs Outflow"): plot_1 = gr.Plot()
+                with gr.TabItem("Analisis Cashflow"): plot_1 = gr.Plot()
                 with gr.TabItem("Tren Pertumbuhan Saldo"): plot_2 = gr.Plot()
-            out_advice = gr.Markdown(elem_classes="advice-box")
+            out_report = gr.Markdown(elem_classes="report-box")
 
-    btn.click(fn=process, inputs=id_in, outputs=[out_status, out_advice, plot_1, plot_2])
+    btn.click(fn=run_archon, inputs=id_in, outputs=[out_status, out_report, plot_1, plot_2])
 
 demo.launch()