sedikit LLM

app.py

@@ -1,112 +1,105 @@
 import pandas as pd
 import numpy as np
 import gradio as gr
-import os
+import plotly.graph_objects as go
 from transformers import pipeline
+from huggingface_hub import InferenceClient
+import os
 
-# --- KONFIGURASI PATH & MODEL ---
-MODEL_PATH = "archon_v1"
-# Kategori wajib sesuai instruksi Fase 2
-ESSENTIAL_CATS = {'groceries', 'utilities', 'transport', 'healthcare', 'education'}
-DISCRETIONARY_CATS = {'restaurant', 'cafe', 'entertainment', 'fashion', 'online_shopping', 'travel'}
+# --- KONFIGURASI PILAR ---
+# Pastikan HF_TOKEN ada di Settings > Secrets di Hugging Face Space kamu
+HF_TOKEN = os.getenv("HF_TOKEN")
+client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.3", token=HF_TOKEN)
 
-class ArchonBankEngine:
+class ArchonFinalEngine:
     def __init__(self):
-        # Pilar 1: AI Classifier (Fase 2)
-        self.classifier = pipeline("text-classification", model=MODEL_PATH, tokenizer=MODEL_PATH)
-        self.load_all_data()
+        # Fase 2: AI Classifier (BERT)
+        self.classifier = pipeline("text-classification", model="archon_v1", tokenizer="archon_v1")
+        self.load_data()
 
-    def load_all_data(self):
-        # FASE 1: DATA FOUNDATION
+    def load_data(self):
+        # Fase 1: Single Source of Truth
         self.df_txn = pd.read_csv('transactions.csv', parse_dates=['date'])
         self.df_cust = pd.read_csv('customers.csv')
         self.df_bal = pd.read_csv('balances_revised.csv', parse_dates=['month'])
         self.df_rep = pd.read_csv('repayments_revised.csv', parse_dates=['due_date'])
-        self.df_off = pd.read_csv('offers.csv')
 
-    def analyze_customer(self, customer_id):
-        # --- PRE-PROCESSING DATA NASABAH ---
+    def get_risk_analysis(self, customer_id):
+        # --- FASE 3 & 4: DETERMINISTIC CALCULATION (Banking Standard) ---
         user_txn = self.df_txn[self.df_txn['customer_id'] == customer_id].copy()
         user_bal = self.df_bal[self.df_bal['customer_id'] == customer_id].sort_values('month')
-        user_rep = self.df_rep[self.df_rep['customer_id'] == customer_id]
-        user_info = self.df_cust[self.df_cust['customer_id'] == customer_id].iloc[0]
-
-        if user_txn.empty: return {"Error": "Data Nasabah Tidak Ditemukan"}
-
-        # --- FASE 2: TRANSACTION INTELLIGENCE (Automasi AI) ---
-        # AI menentukan kategori dari deskripsi transaksi yang ambigu
-        user_txn['merchant_category'] = user_txn['raw_description'].apply(lambda x: self.classifier(x)[0]['label'])
         
-        # Penentuan expense_type sesuai aturan Fase 2
-        def set_expense_type(cat):
-            if any(k in cat.lower() for k in ESSENTIAL_CATS): return 'essential'
-            return 'discretionary'
-        user_txn['expense_type'] = user_txn['merchant_category'].apply(set_expense_type)
-
-        # --- FASE 3 & 4: RISK SCORING (Early Warning System) ---
-        # 1. Expense Ratio (Bulan terakhir)
+        # 1. Expense Ratio Score (30%)
         income = user_txn[user_txn['transaction_type'] == 'credit']['amount'].sum()
         expense = user_txn[user_txn['transaction_type'] == 'debit']['amount'].sum()
         er = expense / income if income > 0 else 1.0
         er_score = 1 if er > 0.8 else (0.5 if er > 0.5 else 0)
 
-        # 2. Balance Trend (Bulan ini vs bulan lalu)
-        if len(user_bal) >= 2:
-            bt = -1 if user_bal.iloc[-1]['avg_balance'] < user_bal.iloc[-2]['avg_balance'] else 1
-        else: bt = 0
-        bt_score = 1 if bt == -1 else 0
-
-        # 3. Overdraft & Missed Payment
+        # 2. Balance Trend Score (20%)
+        bt_score = 1 if len(user_bal) >= 2 and user_bal.iloc[-1]['avg_balance'] < user_bal.iloc[-2]['avg_balance'] else 0
+        
+        # 3. Overdraft Score (20%)
         od_score = 1 if (user_bal['min_balance'] <= 0).any() else 0
-        mp_score = 1 if (user_rep['status'] == 'late').any() else 0
 
-        # HITUNG FINAL RISK SCORE (Bobot Fix: 30%, 20%, 20%, 20%, 10%)
-        risk_score = (0.3 * er_score) + (0.2 * bt_score) + (0.2 * od_score) + (0.2 * mp_score) + 0.1
-        risk_level = "HIGH" if risk_score >= 0.7 else ("MEDIUM" if risk_score >= 0.4 else "LOW")
-
-        # --- FASE 5: NBO ENGINE ---
-        # Menentukan aksi sesuai kriteria Fase 5
-        disc_ratio = user_txn[user_txn['expense_type'] == 'discretionary']['amount'].sum() / expense if expense > 0 else 0
+        # Formula Final Risk Score
+        # risk_score = 0.3 * ER + 0.2 * BT + 0.2 * OD + 0.2 * MP + 0.1 * Vol
+        final_score = (0.3 * er_score) + (0.2 * bt_score) + (0.2 * od_score) + 0.1 
         
-        if risk_level == "HIGH":
-            action = "restructuring_suggestion"
-            reason = "REPAYMENT_RISK_DETECTED"
-        elif risk_level == "MEDIUM" and disc_ratio > 0.4:
-            action = "budgeting_alert"
-            reason = "HIGH_DISCRETIONARY_SPENDING"
-        elif risk_level == "LOW" and disc_ratio <= 0.4:
-            action = "promote_saving"
-            reason = "STABLE_CASHFLOW"
-        else:
-            action = "no_action"
-            reason = "COOLDOWN_ACTIVE"
-
-        # --- FASE 6: EXPLAINABLE SUMMARY ---
-        summary = f"Nasabah memiliki risiko {risk_level}. "
-        if er_score == 1: summary += "Pengeluaran sangat tinggi (>80% pendapatan). "
-        if bt_score == 1: summary += "Tren saldo menurun signifikan. "
-        if mp_score == 1: summary += "Terdeteksi riwayat telat bayar."
-
-        return {
-            "ID Nasabah": customer_id,
-            "Level Risiko": risk_level,
-            "Rekomendasi Aksi": action,
-            "Alasan": reason,
-            "Penjelasan": summary
-        }
-
-# --- INTEGRASI KE DASHBOARD ---
-engine = ArchonBankEngine()
-
-def run_app(cust_id):
-    return engine.analyze_customer(cust_id)
-
-demo = gr.Interface(
-    fn=run_app,
-    inputs=gr.Textbox(label="Input Customer ID (C0001 - C0120)"),
-    outputs="json",
-    title="🛡️ Archon-AI Production Engine",
-    description="Sistem automasi perbankan sesuai instruksi Arahan Pembuatan AI Archon."
-)
+        risk_level = "HIGH" if final_score >= 0.7 else ("MEDIUM" if final_score >= 0.4 else "LOW")
+        
+        return risk_level, final_score, er, user_bal
+
+    def generate_llm_advice(self, risk_level, er_ratio, customer_id):
+        # --- FASE 5: GENERATIVE NBO (Action Layer) ---
+        # Menggunakan LLM agar saran tidak kaku seperti robot
+        prompt = f"""
+        Role: Senior Financial Advisor di Bank Indonesia.
+        Nasabah: {customer_id}
+        Kondisi: Risiko {risk_level}, Rasio Pengeluaran {er_ratio:.2%}.
+        Tugas: Berikan 1 paragraf saran yang natural, empati, dan sangat personal dalam Bahasa Indonesia. 
+        Gunakan gaya bahasa profesional tapi bersahabat. Jangan kaku.
+        """
+        try:
+            response = client.chat_completion(messages=[{"role": "user", "content": prompt}], max_tokens=150)
+            return response.choices[0].message.content
+        except:
+            return "Tetap pantau pengeluaran Anda agar kesehatan finansial terjaga."
+
+    def create_plots(self, user_bal):
+        # --- FASE 6: INSIGHT VISUALIZATION
+        fig = go.Figure()
+        fig.add_trace(go.Scatter(x=user_bal['month'], y=user_bal['avg_balance'], name='Avg Balance', line=dict(color='blue', width=4)))
+        fig.add_trace(go.Bar(x=user_bal['month'], y=user_bal['min_balance'], name='Min Balance', marker_color='orange', opacity=0.6))
+        fig.update_layout(title='Tren Saldo Bulanan', xaxis_title='Bulan', yaxis_title='Jumlah (Rp)', template='plotly_white')
+        return fig
+
+# --- UI LOGIC ---
+engine = ArchonFinalEngine()
+
+def run_archon(cust_id):
+    risk_lv, score, er, bal_data = engine.get_risk_analysis(cust_id)
+    advice = engine.generate_llm_advice(risk_lv, er, cust_id)
+    plot = engine.create_plots(bal_data)
+    
+    summary = {
+        "Risk Level": risk_lv,
+        "Risk Score": round(score, 2),
+        "Expense Ratio": f"{round(er*100, 2)}%"
+    }
+    return summary, advice, plot
+
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🛡️ Archon-AI: Industrial Dashboard")
+    with gr.Row():
+        with gr.Column(scale=1):
+            input_id = gr.Textbox(label="Customer ID", placeholder="C0001")
+            btn = gr.Button("Analyze Customer", variant="primary")
+            out_json = gr.JSON(label="Deterministic Metrics")
+        with gr.Column(scale=2):
+            out_plot = gr.Plot(label="Financial Timeline (Phase 6)")
+    
+    out_advice = gr.Textbox(label="Archon Generative Advice (LLM)", lines=5)
+    
+    btn.click(fn=run_archon, inputs=input_id, outputs=[out_json, out_advice, out_plot])
 
 demo.launch()