Update app.py

app.py

@@ -1985,33 +1985,43 @@ if not df_category.empty:
     # st.markdown(insight_text, unsafe_allow_html=True)
 else:
     st.info("No data available for non-positive issue categories with 100% coverage and positive trend.")
-
-# =================== OBJECTIVE 7 — Insight and Recommendation ===================
 import streamlit as st
 import pandas as pd
-import requests
-
-# =========================
-#  UNIVERSAL LLM CALL
-# =========================
-def llm_call(prompt: str):
-    """Universal call untuk LLM (HF Docker / LM Studio / OpenAI)."""
-    url = "http://localhost:1234/v1/chat/completions"    # UBAH jika perlu
-    payload = {
-        "model": "Qwen2.5-7B-Instruct",                  # UBAH sesuai model
-        "messages": [{"role": "user", "content": prompt}],
-        "temperature": 0.3,
-        "max_tokens": 700
-    }
-    r = requests.post(url, json=payload)
-    r.raise_for_status()
-    return r.json()["choices"][0]["message"]["content"]
+from huggingface_hub import InferenceClient
+
+# ==========================
+# LLM FUNCTION (HuggingFace)
+# ==========================
+def llm_generate_recommendation(insights_text):
+    client = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+
+    prompt = f"""
+    You are an expert Safety & Reliability Agentic AI.
+    Based on the following structured INSIGHT SUMMARY, create:
+    1. Recommended Action (max 2 sentences)
+    2. Risk Mitigation Strategy (max 2 sentences)
+
+    The insights:
+    {insights_text}
+
+    Now generate concise, high-impact:
+    - "recommendation"
+    - "mitigation"
+    Return output in EXACT JSON format:
+    {{
+        "recommendation": "...",
+        "mitigation": "..."
+    }}
+    """
 
+    output = client.text_generation(prompt, max_new_tokens=256, temperature=0.3)
+    return output
 
-# =========================
-#  EXTRACT INSIGHTS
-# =========================
-def extract_agentic_insights(df: pd.DataFrame):
+
+# ==============================================
+# === INSIGHT COMPUTATION FUNCTION (your code)
+# ==============================================
+def extract_agentic_insights_v5(df: pd.DataFrame):
     dev = {
         "lowest_ratio_9_locs": [],
         "obj3a_lowest_div": None,
@@ -2026,179 +2036,168 @@ def extract_agentic_insights(df: pd.DataFrame):
         "obj6_top2_categories": [],
     }
 
-    # === 1. 9 lowest location ratios ===
+    # === 1. 9 locations with lowest finding-to-reporter ratio ===
     if {'nama_lokasi_full', 'creator_nid', 'created_at', 'kode_temuan'}.issubset(df.columns):
-        calc = df.copy()
+        calc = df[['nama_lokasi_full', 'creator_nid', 'created_at', 'kode_temuan']].copy()
         calc['created_at'] = pd.to_datetime(calc['created_at'], errors='coerce')
-        calc = calc.dropna(subset=['created_at', 'creator_nid'])
+        calc = calc.dropna(subset=['created_at', 'nama_lokasi_full', 'creator_nid'])
         calc['bulan'] = calc['created_at'].dt.to_period('M')
-
         monthly = calc.groupby(['nama_lokasi_full', 'bulan']).agg(
             findings=('kode_temuan', 'size'),
             reporters=('creator_nid', 'nunique')
         ).reset_index()
-
         monthly = monthly[monthly['reporters'] > 0]
         monthly['ratio'] = monthly['findings'] / monthly['reporters']
-
         loc_avg = monthly.groupby('nama_lokasi_full')['ratio'].mean()
-        lowest9 = loc_avg.nsmallest(9)
-
-        dev["lowest_ratio_9_locs"] = [(k, round(v, 3)) for k, v in lowest9.items()]
+        lowest_9 = loc_avg.nsmallest(9)
+        dev["lowest_ratio_9_locs"] = [(loc, round(ratio, 3)) for loc, ratio in lowest_9.items()]
 
-    # === 2. Divisions & reporters ===
+    # === 2a Lowest-ratio division ===
     if {'nama', 'creator_nid', 'created_at', 'kode_temuan'}.issubset(df.columns):
-        calc = df.copy()
+        calc = df[['nama', 'creator_nid', 'created_at', 'kode_temuan']].copy()
         calc['bulan'] = pd.to_datetime(calc['created_at']).dt.to_period('M')
-
         agg = calc.groupby(['nama', 'bulan']).agg(
             findings=('kode_temuan', 'size'),
             reporters=('creator_nid', 'nunique')
-        ).reset_index()
-
-        agg = agg[agg['reporters'] > 0]
+        )
+        agg = agg[agg['reporters'] > 0].reset_index()
         agg['ratio'] = agg['findings'] / agg['reporters']
-
         div_ratio = agg.groupby('nama')['ratio'].mean()
         if not div_ratio.empty:
             dev["obj3a_lowest_div"] = (div_ratio.idxmin(), round(div_ratio.min(), 2))
 
-    # Slowest executor
+    # === 2b Slowest executor
     if 'days_to_close' in df.columns:
-        valid = df[df['days_to_close'] >= 0]
-        exec_col = 'nama_pic' if 'nama_pic' in df.columns else 'creator_name'
-
+        valid = df[df['days_to_close'].notna() & (df['days_to_close'] >= 0)]
+        exec_col = 'nama_pic' if 'nama_pic' in valid.columns else 'creator_name'
         if exec_col in valid.columns:
             lead = valid.groupby(exec_col)['days_to_close'].mean()
-            dev["obj3b_slowest_executor"] = (lead.idxmax(), round(lead.max(), 1))
+            if not lead.empty:
+                dev["obj3b_slowest_executor"] = (lead.idxmax(), round(lead.max(), 1))
 
-    # Lowest reporter
+    # === 2c Lowest reporter freq
     if {'creator_name', 'created_at'}.issubset(df.columns):
-        calc = df.copy()
+        calc = df[['creator_name', 'created_at']].copy()
         calc['bulan'] = pd.to_datetime(calc['created_at']).dt.to_period('M')
-        monthly = calc.groupby(['creator_name', 'bulan']).size()
-        avg = monthly.groupby('creator_name').mean()
-        dev["obj3c_lowest_reporter"] = (avg.idxmin(), round(avg.min(), 2))
-
-    # Slowest division (lead time)
-    if {'nama', 'days_to_close'}.issubset(df.columns):
-        lead = df.groupby('nama')['days_to_close'].mean()
-        dev["obj3d_slowest_div"] = (lead.idxmax(), round(lead.max(), 1))
-
-    # === 3. Non positive %
+        monthly = calc.groupby(['creator_name', 'bulan']).size().reset_index(name='count')
+        avg = monthly.groupby('creator_name')['count'].mean()
+        avg = avg[avg > 0]
+        if not avg.empty:
+            dev["obj3c_lowest_reporter"] = (avg.idxmin(), round(avg.min(), 2))
+
+    # === 2d Slowest division resolution
+    if 'days_to_close' in df.columns and 'nama' in df.columns:
+        valid = df[df['days_to_close'].notna() & (df['days_to_close'] >= 0)]
+        if not valid.empty:
+            lead = valid.groupby('nama')['days_to_close'].mean()
+            if not lead.empty:
+                dev["obj3d_slowest_div"] = (lead.idxmax(), round(lead.max(), 1))
+
+    # === 3. Non-Positive findings composition
     if 'temuan_kategori' in df.columns:
         cnt = df['temuan_kategori'].value_counts(normalize=True) * 100
         dev["obj4_unsafe_condition_pct"] = round(cnt.get("Unsafe Condition", 0), 1)
         dev["obj4_unsafe_action_pct"] = round(cnt.get("Unsafe Action", 0), 1)
         dev["obj4_near_miss_pct"] = round(cnt.get("Near Miss", 0), 1)
 
-    # === 4. Quadrants ===
+    # === 4. Quadrants
+    X_LIMIT, Y_LIMIT = 20, 3
     if {'nama', 'created_at', 'days_to_close', 'kode_temuan'}.issubset(df.columns):
         calc = df.copy()
-        calc['created_at'] = pd.to_datetime(calc['created_at'])
-        calc['month'] = calc['created_at'].dt.to_period('M').astype(str)
-
-        monthly_count = calc.groupby(['nama', 'month'])['kode_temuan'].size().reset_index(name='count')
-        avg_count = monthly_count.groupby('nama')['count'].mean().reset_index(name='Finding Count')
-        avg_lead = calc.groupby('nama')['days_to_close'].mean().reset_index(name='Avg Lead Time')
-
-        m = avg_count.merge(avg_lead, on='nama')
-        X, Y = 20, 3
-
-        for _, r in m.iterrows():
-            if r['Finding Count'] >= X and r['Avg Lead Time'] >= Y:
+        calc['created_at'] = pd.to_datetime(calc['created_at'], errors='coerce')
+        calc = calc.assign(month=calc['created_at'].dt.to_period('M').astype(str))
+        monthly_counts = calc.groupby(['nama', 'month'])['kode_temuan'].nunique().reset_index()
+        avg_count = monthly_counts.groupby('nama')['kode_temuan'].mean().reset_index(name='Finding Count')
+        leadtime = calc.groupby('nama')['days_to_close'].mean().reset_index(name='Avg Lead Time')
+        mat = avg_count.merge(leadtime, on='nama', how='left').fillna(0)
+        for _, r in mat.iterrows():
+            if r['Finding Count'] >= X_LIMIT and r['Avg Lead Time'] >= Y_LIMIT:
                 dev["obj5_q1_divs"].append(r['nama'])
-            elif r['Finding Count'] < X and r['Avg Lead Time'] >= Y:
+            elif r['Finding Count'] < X_LIMIT and r['Avg Lead Time'] >= Y_LIMIT:
                 dev["obj5_q2_divs"].append(r['nama'])
 
-    # === 5. top 2 category ===
+    # === 5. Top categories
     if {'kategori', 'temuan_kategori', 'created_at'}.issubset(df.columns):
-        nonpos = df[df['temuan_kategori'] != "Positive"]
-        start = nonpos['created_at'].min().to_period('M')
-        end = nonpos['created_at'].max().to_period('M')
-        n_months = len(pd.period_range(start, end, freq='M'))
-        cat_avg = nonpos.groupby('kategori').size() / n_months
-        cat_avg = cat_avg.sort_values(ascending=False).head(2)
-        dev["obj6_top2_categories"] = [(k, round(v, 1)) for k, v in cat_avg.items()]
+        nonpos = df[df['temuan_kategori'] != 'Positive']
+        if not nonpos.empty:
+            start = nonpos['created_at'].min().to_period('M')
+            end = nonpos['created_at'].max().to_period('M')
+            n_months = len(pd.period_range(start=start, end=end, freq='M'))
+            cat_avg = (nonpos.groupby('kategori').size() / n_months).sort_values(ascending=False).head(2)
+            dev["obj6_top2_categories"] = [(cat, round(val, 1)) for cat, val in cat_avg.items()]
 
     return dev
 
 
-# =========================
-#  RENDER + LLM GENERATION
-# =========================
+# ==========================
+# ===== MAIN APP ===========
+# ==========================
 
-dev = extract_agentic_insights(df_filtered)
+st.markdown("<h3 class='section-title'>OBJECTIVE 7 — Insight and Recommendation (LLM powered)</h3>", unsafe_allow_html=True)
 
-# ======== BUILD TEXT FOR LLM ========
-prompt = f"""
-You are an advanced Safety Analytics LLM.
+dev = extract_agentic_insights_v5(df_filtered)
 
-Given the following structured insights from real safety operational data:
+# Build INSIGHT SUMMARY as text for LLM
+summary_parts = []
 
-{dev}
+if dev["lowest_ratio_9_locs"]:
+    summary_parts.append(f"Lowest-ratio locations: {dev['lowest_ratio_9_locs']}")
 
-Your tasks:
+if dev["obj3a_lowest_div"]:
+    summary_parts.append(f"Lowest performing division: {dev['obj3a_lowest_div']}")
 
-1. Write a concise **Insight Summary** (max 6 bullets). Use corporate tone and highlight anomalies.
-2. Generate **5 Recommended Actions**, each 1–2 sentences.
-3. Generate **5 Risk Mitigation Strategies**, each paired to each recommendation.
+if dev["obj3b_slowest_executor"]:
+    summary_parts.append(f"Slowest executor: {dev['obj3b_slowest_executor']}")
 
-Return output in this JSON structure ONLY:
+if dev["obj3c_lowest_reporter"]:
+    summary_parts.append(f"Least active reporter: {dev['obj3c_lowest_reporter']}")
 
-{{
-"insight_summary": "...",
-"recommendations": ["...", "...", "...", "...", "..."],
-"mitigations": ["...", "...", "...", "...", "..."]
-}}
-"""
+if dev["obj3d_slowest_div"]:
+    summary_parts.append(f"Slowest division resolution: {dev['obj3d_slowest_div']}")
 
-llm_output = llm_call(prompt)
+uc, ua, nm = dev["obj4_unsafe_condition_pct"], dev["obj4_unsafe_action_pct"], dev["obj4_near_miss_pct"]
+summary_parts.append(f"Non-Positive: UnsafeCondition={uc}%, UnsafeAction={ua}%, NearMiss={nm}%")
 
-import json
-out = json.loads(llm_output)
+summary_parts.append(f"Quadrant I: {dev['obj5_q1_divs']}")
+summary_parts.append(f"Quadrant II: {dev['obj5_q2_divs']}")
+
+if dev["obj6_top2_categories"]:
+    summary_parts.append(f"Top non-positive categories: {dev['obj6_top2_categories']}")
 
-# ----------------------------
-#  STREAMLIT RENDERING
-# ----------------------------
-st.markdown("<h3 class='section-title'>OBJECTIVE 7 — Insight and Recommendation</h3>", unsafe_allow_html=True)
+insight_summary_text = "\n".join(summary_parts)
 
-# Insight card
+# Call LLM to generate recommendation + mitigation
+llm_json = llm_generate_recommendation(insight_summary_text)
+
+# Try to parse JSON
+import json
+try:
+    llm_output = json.loads(llm_json)
+    recommendation = llm_output["recommendation"]
+    mitigation = llm_output["mitigation"]
+except:
+    recommendation = "LLM output not valid JSON."
+    mitigation = "-"
+
+# Render
 st.markdown(
     f"""
-    <div style="background:#f8f9fa; border-left:4px solid #003DA5; padding:16px; border-radius:4px;">
-        <h4 style="margin:0;color:#003DA5;">Insight Summary (LLM Generated)</h4>
-        <p style="line-height:1.6;">{out['insight_summary'].replace("\n", "<br>")}</p>
-    </div>""",unsafe_allow_html=True)
-
-# Recommendations + Mitigations table
-rows = ""
-for i in range(5):
-    rows += f"""
-    <tr>
-        <td style='text-align:center; font-weight:bold;'>{i+1}</td>
-        <td style='padding:8px;'>{out['recommendations'][i]}</td>
-        <td style='padding:8px;'>{out['mitigations'][i]}</td>
-    </tr>
-    """
+    <div style="padding:15px; background:#eef3ff; border-left:4px solid #003DA5;">
+        <h4>Insight Summary</h4>
+        <pre style="white-space: pre-wrap;">{insight_summary_text}</pre>
+    </div>
+    """,
+    unsafe_allow_html=True,
+)
 
 st.markdown(
     f"""
-    <div style="background:#e8f5e9; border-left:4px solid #4CAF50; padding:16px; border-radius:4px;">
-        <h4 style="margin:0;color:#2E7D32;">Recommended Actions & Agentic Risk Mitigation (LLM)</h4>
-        <table style="width:100%; border-collapse:collapse; margin-top:12px;">
-            <thead>
-                <tr style="background:#d4efdf;">
-                    <th>#</th>
-                    <th>Recommended Action</th>
-                    <th>Risk Mitigation</th>
-                </tr>
-            </thead>
-            <tbody>
-                {rows}
-            </tbody>
-        </table>
+    <div style="padding:15px; background:#e8f5e9; border-left:4px solid #4CAF50; margin-top:20px;">
+        <h4>LLM Recommended Action</h4>
+        <p>{recommendation}</p>
+        <h4>LLM Risk Mitigation</h4>
+        <p>{mitigation}</p>
     </div>
     """,
-    unsafe_allow_html=True
+    unsafe_allow_html=True,
 )