Update app.py

app.py

@@ -2,6 +2,9 @@ import gradio as gr
 import joblib
 import pandas as pd
 import numpy as np
+import os
+from groq import Groq
+
 from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer
 from reportlab.lib.styles import ParagraphStyle, getSampleStyleSheet
 from reportlab.lib import colors
@@ -12,10 +15,54 @@ from reportlab.platypus import TableStyle
 # Load model
 model = joblib.load("typhoidguard_model_reduced.pkl")
 
-# Risk & Agentic Logic
+# Groq Client (API key from HF Secrets)
+client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
+
+# ===============================
+# LLM REPORT FUNCTION
+# ===============================
+def generate_llm_report(patient_info, prob_pct, risk_label, recommendation, factors):
+
+    prompt = f"""
+You are a responsible medical AI assistant helping rural healthcare units in Pakistan.
+
+Patient Details:
+{patient_info}
+
+Predicted Risk of Treatment Failure: {prob_pct}%
+Risk Category: {risk_label}
+
+Clinical Risk Factors:
+{", ".join(factors) if factors else "None"}
+
+Agent Recommendation:
+{recommendation}
+
+Generate a structured detailed report including:
+1. Clinical Interpretation
+2. Possible Complications
+3. Immediate Precautions
+4. Preventative Measures
+5. Follow-up Recommendations
+
+Use clear bullet points.
+End with a reminder that final decisions must be made by a qualified doctor.
+"""
+
+    response = client.chat.completions.create(
+        model="llama-3.3-70b-versatile",
+        messages=[{"role": "user", "content": prompt}],
+        temperature=0.4,
+    )
+
+    return response.choices[0].message.content
+
+
+# ===============================
+# Risk & Agentic Logic (UNCHANGED)
+# ===============================
 def predict_and_generate(name, age, gender, platelets, hb, duration, calcium, potassium):
 
-    # 1️⃣ Prepare data with exact column names
     data = pd.DataFrame([{
         "Platelet Count": platelets,
         "Age": age,
@@ -26,24 +73,18 @@ def predict_and_generate(name, age, gender, platelets, hb, duration, calcium, po
         "Gender": gender
     }])
 
-    # 2️⃣ Transform through preprocessor
     X_processed = model.named_steps["preprocessor"].transform(data)
-
-    # 3️⃣ Model probability
     prob = model.named_steps["classifier"].predict_proba(X_processed)[0][1]
 
-    # 4️⃣ Clinical thresholds
     low_platelets = platelets < 150000
     low_hb = hb < 10
     prolonged_treatment = duration > 7
 
-    # 5️⃣ Adjust probability for extreme lab values
     if low_platelets or low_hb:
-        prob_pct = 90.0  # clinically override
+        prob_pct = 90.0
     else:
         prob_pct = round(prob * 100, 2)
 
-    # Default risk label based on probability
     if prob > 0.65:
         risk_label = "🔴 HIGH RISK"
         color = "#d62828"
@@ -54,7 +95,6 @@ def predict_and_generate(name, age, gender, platelets, hb, duration, calcium, po
         risk_label = "🟢 LOW RISK"
         color = "#2a9d8f"
 
-    # Override risk for critical lab values
     if low_platelets or low_hb:
         risk_label = "🔴 HIGH RISK"
         color = "#d62828"
@@ -64,7 +104,6 @@ def predict_and_generate(name, age, gender, platelets, hb, duration, calcium, po
     else:
         recommendation = "Continue standard treatment protocol."
 
-    # 6️⃣ Risk factors list
     factors = []
     if low_platelets:
         factors.append("Low Platelet Count")
@@ -77,142 +116,78 @@ def predict_and_generate(name, age, gender, platelets, hb, duration, calcium, po
     if potassium < 3.5 or potassium > 5.0:
         factors.append("Abnormal Potassium")
 
-    # 7️⃣ Construct HTML
+    # -------- LLM CALL --------
+    patient_info = f"""
+Age: {age}
+Gender: {gender}
+Platelets: {platelets}
+Hemoglobin: {hb}
+Treatment Duration: {duration}
+Calcium: {calcium}
+Potassium: {potassium}
+"""
+
+    llm_report = generate_llm_report(
+        patient_info, prob_pct, risk_label, recommendation, factors
+    )
+
+    # -------- HTML OUTPUT --------
     explanation = f"""
     <div style="font-family:Arial; padding:15px;">
         <h2 style="color:{color};">Patient: {name}</h2>
         <h3>Risk of Treatment Failure: {prob_pct}%</h3>
         <h3>Status: {risk_label}</h3>
+
         <h4>Clinical Risk Factors:</h4>
         <ul>
-            {''.join([f"<li>{f}</li>" for f in factors])}
+            {''.join([f"<li>{f}</li>" for f in factors]) if factors else "<li>None</li>"}
         </ul>
+
         <h4>Agent Recommendation:</h4>
-        <p>{recommendation}</p>
+        <p><b>{recommendation}</b></p>
+
+        <hr>
+
+        <h3 style="color:#003366;">🧠 Detailed AI Report & Recommendations</h3>
+        <div style="background:#f8f9fa; padding:15px; border-radius:10px; line-height:1.6;">
+            {llm_report.replace("\n", "<br>")}
+        </div>
     </div>
     """
 
     return explanation
 
 
-# PDF Generation
-# PDF Generation (with color coding & disclaimer)
+# ===============================
+# PDF Generation (LLM Included)
+# ===============================
 def generate_pdf(name, age, gender, platelets, hb, duration, calcium, potassium):
 
-    # 1️⃣ Prepare data for model & agentic logic
-    data = pd.DataFrame([{
-        "Platelet Count": platelets,
-        "Age": age,
-        "Hemoglobin (g/dL)": hb,
-        "Treatment Duration": duration,
-        "Calcium (mg/dL)": calcium,
-        "Potassium (mmol/L)": potassium,
-        "Gender": gender
-    }])
-
-    X_processed = model.named_steps["preprocessor"].transform(data)
-    prob = model.named_steps["classifier"].predict_proba(X_processed)[0][1]
+    html_output = predict_and_generate(name, age, gender, platelets, hb, duration, calcium, potassium)
 
-    low_platelets = platelets < 150000
-    low_hb = hb < 10
-    prolonged_treatment = duration > 7
-
-    # Adjust probability for critical lab values
-    if low_platelets or low_hb:
-        prob_pct = 95.0
-    else:
-        prob_pct = round(prob * 100, 2)
-
-    # Determine status and recommendation
-    if prob > 0.65:
-        risk_label = "HIGH RISK"
-        color = colors.red
-    elif prob >= 0.35:
-        risk_label = "MODERATE RISK"
-        color = colors.orange
-    else:
-        risk_label = "LOW RISK"
-        color = colors.green
-
-    if low_platelets or low_hb:
-        risk_label = "HIGH RISK"
-        color = colors.red
-        recommendation = "Immediate referral to secondary care facility recommended. Close monitoring required."
-    elif prob >= 0.35:
-        recommendation = "Close monitoring advised. Repeat blood tests within 48 hours."
-    else:
-        recommendation = "Continue standard treatment protocol."
-
-    # Clinical risk factors
-    factors = []
-    if low_platelets:
-        factors.append("Low Platelet Count")
-    if low_hb:
-        factors.append("Low Hemoglobin")
-    if prolonged_treatment:
-        factors.append("Prolonged Treatment Duration")
-    if calcium < 8.5:
-        factors.append("Low Calcium")
-    if potassium < 3.5 or potassium > 5.0:
-        factors.append("Abnormal Potassium")
-
-    # 2️⃣ Generate PDF
     file_path = f"{name}_Typhoid_Report.pdf"
     doc = SimpleDocTemplate(file_path)
     elements = []
     styles = getSampleStyleSheet()
 
-    # Title
     elements.append(Paragraph("<b>TyphoidGuard AI Clinical Report</b>", styles["Title"]))
     elements.append(Spacer(1, 0.3 * inch))
 
-    # Patient Info Table
-    patient_data = [
-        ["Patient Name", name],
-        ["Age", age],
-        ["Gender", gender],
-        ["Platelet Count", platelets],
-        ["Hemoglobin (g/dL)", hb],
-        ["Treatment Duration (Days)", duration],
-        ["Calcium (mg/dL)", calcium],
-        ["Potassium (mmol/L)", potassium],
-        ["Risk of Treatment Failure (%)", prob_pct],
-        ["Status", risk_label],
-    ]
-
-    table = Table(patient_data, colWidths=[200, 250])
-    table.setStyle(TableStyle([
-        ('BACKGROUND', (0,0), (-1,0), colors.lightblue),
-        ('GRID', (0,0), (-1,-1), 1, colors.grey),
-        ('FONTNAME', (0,0), (-1,0), 'Helvetica-Bold'),
-        ('VALIGN', (0,0), (-1,-1), 'MIDDLE'),
-        # Color coding for risk status
-        ('TEXTCOLOR', (-1,-1), (-1,-1), color)
-    ]))
-    elements.append(table)
-    elements.append(Spacer(1, 0.3 * inch))
-
-    # Clinical Risk Factors
-    if factors:
-        elements.append(Paragraph("<b>Clinical Risk Factors:</b>", styles["Heading3"]))
-        for f in factors:
-            elements.append(Paragraph(f"- {f}", styles["Normal"]))
-        elements.append(Spacer(1, 0.2 * inch))
-
-    # Agent Recommendation
-    elements.append(Paragraph("<b>Agent Recommendation:</b>", styles["Heading3"]))
-    elements.append(Paragraph(recommendation, styles["Normal"]))
+    elements.append(Paragraph(html_output.replace("<br>", "<br/>"), styles["Normal"]))
     elements.append(Spacer(1, 0.3 * inch))
 
-    # Footer
-    elements.append(Paragraph("Generated under HEC Generative AI Training Hackathon", styles["Normal"]))
-    elements.append(Paragraph("Team Chinar AI - AJK", styles["Normal"]))
-    elements.append(Spacer(1, 0.2 * inch))
-    elements.append(Paragraph("<i>Disclaimer: This report is AI-generated. Please consult a qualified doctor for clinical decisions.</i>", styles["Normal"]))
+    elements.append(Paragraph(
+        "<i>Disclaimer: This report is AI-generated. Please consult a qualified doctor before making clinical decisions.</i>",
+        styles["Normal"]
+    ))
 
     doc.build(elements)
     return file_path
+
+
+# ===============================
 # Interface
+# ===============================
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
 
     gr.Markdown("""
@@ -237,7 +212,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
 
     output = gr.HTML()
 
-    predict_btn = gr.Button("🔍 Analyze Patient")
+    predict_btn = gr.Button("🧠 Detailed Report and Recommendations")
     pdf_btn = gr.Button("📄 Generate PDF Report")
 
     predict_btn.click(