Update src/streamlit_app.py

src/streamlit_app.py

@@ -761,5 +761,299 @@ with tabs[8]:
     else:
         st.info("No logs yet — run AutoML once.")
 
+
+# ----- Smart Advisor tab (Role-based Insights)
+with tabs[9]:
+    st.subheader(" Smart Advisor — Role-Based Insights")
+
+    # -------------------------
+    # 1. Role hierarchy and descriptions
+    # -------------------------
+    roles = {
+        # --- Ground-level ---
+        "Furnace Operator": "Runs daily EAF heats, manages electrodes, slag foaming, and tap timing.",
+        "Shift Engineer": "Coordinates furnace, casting, and maintenance operations during the shift.",
+        "Process Metallurgist": "Optimizes chemistry, refining, and metallurgical balance across heats.",
+        "Maintenance Engineer": "Monitors vibration, bearings, and schedules preventive maintenance.",
+        "Quality Engineer": "Tracks billet surface, composition, and defect rates from casting to rolling.",
+
+        # --- Mid-level / Plant ---
+        "Energy Manager": "Analyzes power, load factor, and energy cost per ton of steel.",
+        "Production Head": "Supervises throughput, yield, and adherence to shift-level production targets.",
+        "Reliability Manager": "Oversees equipment reliability, predictive maintenance, and downtime prevention.",
+        "Chief Process Engineer": "Links metallurgical parameters to standard operating conditions.",
+        "Process Optimization Head (PP&C)": "Balances yield, power, and reliability across EAF, caster, and rolling units.",
+        "Chief General Manager – PP&C": "Oversees planning, process, and control at plant level — coordinating all shops for optimal energy, yield, and reliability.",
+        "Deputy General Manager (Operations)": "Supervises multi-shop coordination, productivity, and manpower scheduling.",
+        "Plant Head": "Oversees plant-wide KPIs — production, energy, quality, and modernization progress.",
+
+        # --- Strategic / Corporate ---
+        "Executive Director (Works)": "Integrates operations, people, and safety across all plants.",
+        "Chief Operating Officer (COO)": "Ensures alignment between production efficiency and business goals.",
+        "Chief Sustainability Officer (CSO)": "Monitors CO₂ intensity, waste recovery, and environmental compliance.",
+        "Chief Financial Officer (CFO)": "Links operational performance to cost efficiency and ROI.",
+        "Chief Executive Officer (CEO)": "Focuses on long-term performance, modernization, and shareholder impact."
+    }
+
+    # -------------------------
+    # 2. Role-specific reasoning prompts for LLM
+    # -------------------------
+    role_prompts = {
+        "Furnace Operator": """
+            You are the EAF furnace operator responsible for maintaining a stable arc and safe melting.
+            Translate model recommendations into clear, actionable controls: electrode movement, oxygen flow,
+            slag foaming, or power adjustment. Focus on operational safety and tap timing.
+        """,
+        "Shift Engineer": """
+            You are the shift engineer overseeing melting, casting, and maintenance coordination.
+            Interpret model insights for operational actions — mention if inter-shop coordination is required.
+        """,
+        "Process Metallurgist": """
+            You are the process metallurgist. Evaluate the data-driven SHAP patterns to interpret metallurgical
+            balance, oxidation behavior, and refining efficiency. Suggest chemistry or process tuning.
+        """,
+        "Maintenance Engineer": """
+            You are the maintenance engineer responsible for reliability. Identify potential failure risks
+            (e.g., vibration anomalies, overheating, current imbalance) and propose proactive checks.
+        """,
+        "Quality Engineer": """
+            You are the quality engineer monitoring casting and rolling outcomes. Translate process variables
+            into expected surface or composition quality impacts and preventive measures.
+        """,
+        "Energy Manager": """
+            You are the energy manager. Interpret how SHAP signals influence energy per ton and power factor.
+            Quantify efficiency deviations and suggest scheduling or load adjustments.
+        """,
+        "Production Head": """
+            You are the production head tracking yield and throughput. Connect SHAP insights to bottlenecks
+            in productivity, heat timing, or equipment utilization. Suggest optimization steps.
+        """,
+        "Reliability Manager": """
+            You are the reliability manager. Evaluate if process trends suggest equipment stress, overheating,
+            or wear. Recommend intervention plans and projected downtime avoidance.
+        """,
+        "Chief Process Engineer": """
+            You are the chief process engineer. Convert SHAP outputs into process standardization insights.
+            Flag anomalies that require SOP review and coordinate with metallurgical and control teams.
+        """,
+        "Process Optimization Head (PP&C)": """
+            You are the Process Optimization Head in PP&C. Assess SHAP signals across multiple units to improve
+            system-level yield, energy, and reliability. Recommend balanced actions and inter-shop alignment.
+        """,
+        "Chief General Manager – PP&C": """
+            You are the Chief General Manager (PP&C) responsible for overall plant coordination,
+            planning, process control, and modernization. Interpret model insights as if briefing
+            senior management and section heads before a shift review.
+
+            Your response must:
+            - Translate technical terms into operational themes (e.g., “arc instability”)
+            - Identify cross-functional effects (EAF ↔ Caster ↔ Rolling)
+            - Suggest coordination steps (maintenance, power, metallurgist)
+            - Conclude with KPI or strategic impact (yield, energy, reliability)
+            - If any data pattern seems implausible, mention it and propose review.
+        """,
+        "Deputy General Manager (Operations)": """
+            You are the DGM (Operations). Summarize SHAP-derived insights into actionable instructions
+            for shop heads. Emphasize throughput, manpower planning, and heat plan adherence.
+        """,
+        "Plant Head": """
+            You are the Plant Head. Translate technical findings into KPI performance trends and upcoming
+            operational risks. Recommend cross-departmental actions and expected impact on production targets.
+        """,
+        "Executive Director (Works)": """
+            You are the Executive Director (Works). Summarize how the plant is performing overall and where
+            immediate leadership attention is required. Use a governance-level tone, referencing key KPIs.
+        """,
+        "Chief Operating Officer (COO)": """
+            You are the COO. Interpret model insights at a strategic level — efficiency, tonnage, cost, reliability.
+            Highlight systemic improvements, risk areas, and financial implications across plants.
+        """,
+        "Chief Sustainability Officer (CSO)": """
+            You are the CSO. Relate operational insights to environmental impact, carbon efficiency,
+            and sustainability metrics. Quantify potential emission reduction.
+        """,
+        "Chief Financial Officer (CFO)": """
+            You are the CFO. Interpret operational SHAP findings in terms of cost efficiency, asset utilization,
+            and ROI. Provide an executive financial perspective on potential savings or risks.
+        """,
+        "Chief Executive Officer (CEO)": """
+            You are the CEO of a major integrated steel producer.
+            Provide a concise narrative (2–3 paragraphs) summarizing plant performance trends,
+            operational risks, and opportunities — linking them to strategic goals in the annual report:
+            productivity, sustainability, cost leadership, and modernization.
+        """
+    }
+
+    # -------------------------
+    # 3. Role selection and contextual info
+    # -------------------------
+    role = st.selectbox("Select Your Role", list(roles.keys()), index=10)
+    st.caption(f" Context: {roles[role]}")
+
+    if "recs" not in locals() or not isinstance(recs, list) or not recs:
+        st.warning("Please run the AutoML + SHAP step first to generate recommendations.")
+    else:
+        if st.button("Generate Role-Based Advisory"):
+            HF_TOKEN = os.getenv("HF_TOKEN")
+            if not HF_TOKEN:
+                st.error("HF_TOKEN not found. Please set it as an environment variable or in secrets.toml.")
+            else:
+                import requests, textwrap
+
+                API_URL = "https://router.huggingface.co/v1/chat/completions"
+                headers = {"Authorization": f"Bearer {HF_TOKEN}", "Content-Type": "application/json"}
+
+                # Add CGM-style reasoning context for cross-functional roles
+                if role in ["Chief General Manager – PP&C", "Process Optimization Head (PP&C)", "Plant Head"]:
+                    reasoning_context = """
+                    Think like a systems integrator balancing EAF, caster, and rolling mill performance.
+                    Evaluate interdependencies and recommend coordinated actions across departments.
+                    """
+                elif role in ["COO", "CFO", "CEO"]:
+                    reasoning_context = """
+                    Think strategically. Connect operational drivers to business KPIs,
+                    and quantify financial or sustainability implications.
+                    """
+                else:
+                    reasoning_context = ""
+
+                # Build final prompt
+                prompt = textwrap.dedent(f"""
+                    Role: {role}
+                    Use case: {use_case}
+                    Target variable: {target}
+                    Ensemble model confidence (R²): {final_r2:.3f}
+
+                    {reasoning_context}
+
+                    Model-derived recommendations:
+                    {json.dumps(recs, indent=2)}
+
+                    {role_prompts.get(role, "Provide a professional metallurgical advisory summary.")}
+                    
+                    Your response should cover:
+                    1. What’s happening (interpreted simply)
+                    2. What should be done
+                    3. What outcomes to expect and why
+                """)
+
+                payload = {
+                    "model": "meta-llama/Meta-Llama-3-8B-Instruct",
+                    "messages": [
+                        {"role": "system", "content": "You are a multi-role metallurgical advisor connecting data to human decisions."},
+                        {"role": "user", "content": prompt}
+                    ],
+                    "temperature": 0.4,
+                    "max_tokens": 350,
+                }
+
+                with st.spinner(f"Generating role-based advisory for {role}..."):
+                    resp = requests.post(API_URL, headers=headers, json=payload, timeout=90)
+                    if resp.status_code == 200:
+                        data = resp.json()
+                        msg = (
+                            data.get("choices", [{}])[0]
+                            .get("message", {})
+                            .get("content", "")
+                            .strip()
+                        )
+                        if msg:
+                            st.markdown(f"### Advisory for {role}")
+                            st.info(msg)
+                            # ---- Dynamic Data-Driven Highlights ----
+                            if role in ["Chief General Manager – PP&C", "Plant Head", "Process Optimization Head (PP&C)"]:
+                                st.markdown("#### 🔍 Shift Highlights — Data-Driven Summary")
+                        
+                                try:
+                                    # 1️⃣ Latest data snapshot (simulates "current shift")
+                                    latest_df = df.tail(500).copy()  # last 500 rows = current shift snapshot
+                        
+                                    # 2️⃣ Basic KPIs computed from live data
+                                    furnace_temp_mean = latest_df["furnace_temp"].mean()
+                                    furnace_temp_std = latest_df["furnace_temp"].std()
+                                    energy_eff_mean = latest_df["energy_efficiency"].mean()
+                                    yield_mean = latest_df["yield_ratio"].mean()
+                                    downtime_proxy = np.mean(latest_df["refractory_limit_flag"]) * 8  # hours/shift (proxy)
+                        
+                                    # 3️⃣ Compare vs historical baseline (previous 500 rows)
+                                    if len(df) > 1000:
+                                        prev_df = df.tail(1000).head(500)
+                                        delta_temp = ((furnace_temp_mean - prev_df["furnace_temp"].mean()) /
+                                                      prev_df["furnace_temp"].mean()) * 100
+                                        delta_eff = ((energy_eff_mean - prev_df["energy_efficiency"].mean()) /
+                                                     prev_df["energy_efficiency"].mean()) * 100
+                                        delta_yield = ((yield_mean - prev_df["yield_ratio"].mean()) /
+                                                       prev_df["yield_ratio"].mean()) * 100
+                                    else:
+                                        delta_temp, delta_eff, delta_yield = 0, 0, 0
+                        
+                                    # 4️⃣ Qualitative interpretation
+                                    def trend_symbol(val):
+                                        if val > 0.5:
+                                            return f"↑ +{val:.2f}%"
+                                        elif val < -0.5:
+                                            return f"↓ {val:.2f}%"
+                                        else:
+                                            return f"→ {val:.2f}%"
+                        
+                                    # 5️⃣ Build structured table
+                                    highlights = pd.DataFrame([
+                                        ["Furnace Temp Stability",
+                                         "Stable" if furnace_temp_std < 50 else "Fluctuating",
+                                         f"Avg: {furnace_temp_mean:.1f}°C ({trend_symbol(delta_temp)})"],
+                                        ["Energy Efficiency",
+                                         "Improved" if delta_eff > 0 else "Declined",
+                                         f"{energy_eff_mean:.4f} ({trend_symbol(delta_eff)})"],
+                                        ["Yield Ratio",
+                                         "Nominal" if abs(delta_yield) < 1 else ("↑" if delta_yield > 0 else "↓"),
+                                         f"{yield_mean*100:.2f}% ({trend_symbol(delta_yield)})"],
+                                        ["Refractory Limit Flag",
+                                         "Within Safe Limit" if downtime_proxy < 1 else "Check Lining",
+                                         f"Active Alerts: {downtime_proxy:.1f}/shift"]
+                                    ], columns=["Parameter", "Status", "Observation"])
+                        
+                                    st.dataframe(highlights, use_container_width=True)
+                                    st.caption("Derived from live dataset trends (last 500 vs previous 500 rows).")
+                        
+                                    # 6️⃣ Optional: Link to SHAP recs for validation
+                                    if isinstance(recs, list) and recs:
+                                        st.markdown("#### 🧭 Cross-Verification with SHAP Insights")
+                                        matches = [r for r in recs if any(k in r for k in ["furnace", "energy", "yield", "slag", "power"])]
+                                        if matches:
+                                            st.info("Aligned SHAP Recommendations:\n\n- " + "\n- ".join(matches))
+                                        else:
+                                            st.warning("No direct SHAP alignment found — potential anomaly or unseen pattern.")
+                                except Exception as e:
+                                    st.warning(f"Shift Highlights unavailable: {e}")
+
+                        else:
+                            st.warning(f"Empty response.\nRaw: {data}")
+                    else:
+                        st.error(f"HF API error {resp.status_code}: {resp.text}")
+
+    # -------------------------
+    # 4. Optional role-based KPIs
+    # -------------------------
+    if role == "Chief General Manager – PP&C":
+        col1, col2, col3 = st.columns(3)
+        col1.metric("Plant Yield (Rolling 24h)", "96.8%", "↑0.7% vs yesterday")
+        col2.metric("Energy per ton", "4.92 MWh/t", "↓2.3% week-on-week")
+        col3.metric("Unplanned Downtime", "3.1 hrs", "↓1.2 hrs")
+        st.caption("KPIs aligned with PP&C Balanced Scorecard — Yield • Energy • Reliability")
+
+    elif role in ["CEO", "COO"]:
+        col1, col2, col3 = st.columns(3)
+        col1.metric("EBITDA per ton", "₹7,420", "↑3.1% QoQ")
+        col2.metric("CO₂ Intensity", "1.79 tCO₂/t", "↓2.4% YoY")
+        col3.metric("Modernization CapEx", "₹122 Cr", "On track")
+        st.caption("Strategic alignment: cost leadership • sustainability • modernization")
+
+    elif role in ["Furnace Operator", "Shift Engineer"]:
+        col1, col2, col3 = st.columns(3)
+        col1.metric("Furnace Temp", f"{df['furnace_temp'].iloc[-1]:.1f} °C")
+        col2.metric("Arc Power", f"{df['arc_power'].iloc[-1]:.0f} kW")
+        col3.metric("Power Factor", f"{df['power_factor'].iloc[-1]:.2f}")
+        st.caption("Live operational parameters — monitor stability and foaming balance.")
+
 st.markdown("---")
 st.markdown("**Note:** Synthetic demo dataset for educational use only. Real deployment requires plant data, NDA, and safety validation.")