app.py

import streamlit as st
import torch
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

from transformers import pipeline
pipe = pipeline("text-classification", model="distilbert-base-uncased")  # 67MB vs BERT's 440MB

classifier = pipeline("text-classification", model="distilbert-base-uncased-finetuned-sst-2-english")

import time
from datetime import datetime

@st.cache_data
def load_model():
    return pipeline("text-classification", model="distilbert-base-uncased")  # Cached after first run


# from transformers import pipeline
# pipe = pipeline('text-generation', model='gpt2', device=0 if torch.cuda.is_available() else -1)

# from transformers import pipeline
# classifier = pipeline("text-classification", model="distilbert-base-uncased", device="cpu")  # Faster init

#########

# import os
# from fastapi import FastAPI
# from fastapi.middleware.wsgi import WSGIMiddleware
# from streamlit.web.server import Server

# app = FastAPI()
# app.mount("/", WSGIMiddleware(Server._get_app().wsgi_app))

# @app.get("/health")
# def health_check():
#     return {"status": "healthy"}

# if os.getenv("STREAMLIT_CLOUD"):
#     import uvicorn
#     uvicorn.run(app, host="0.0.0.0", port=8501)

# @st.cache_resource(ttl=3600, max_entries=3)
# def load_model():
#  return pipeline('text-generation', model='gpt2')  # Use smaller models

########## health check for server

# 1. APP CONFIGURATION ================================================
st.set_page_config(
    page_title="Counselor Guidance Assistant",
    page_icon="🧠",
    layout="centered",
    initial_sidebar_state="expanded"
)

# 2. CUSTOM STYLING ==================================================
st.markdown("""
<style>
 :root {
  --primary_clr: #43a573;
  --secondary: #f8f9fa;
  --font_clr: #1a1a1a;
  --bg_clr: #e9f5ef;
  --bg_clr_g2: #d9ede3;
  --bg_clr_g3: #1b422e;
  --border_clr: #1b422e;


  /* Dark Mode Overrides */
  --dark-primary: #43a573;
  --dark-secondary: #2d3748;
  --dark-font: #e2e8f0;
  --dark-bg: #1a202c;
  --dark-bg2: #2d3748;
  --dark-bg3: #38a169;
  --dark-border: #4a5568;
 }



 * {
  box-sizing: border-box;
 }

 html,body {
  font-family: sans-serif;
  line-height: 1.5;
  -webkit-text-size-adjust: 100%;
  -ms-text-size-adjust: 100%;
  -ms-overflow-style: scrollbar;
  -webkit-tap-highlight-color: transparent;
  margin:0;
  height: 100%;
  color: var(--font_clr);
  background-color: var(--bg_clr)
 }

 @-ms-viewport {
 width: device-width;
 }

 .stAppHeader, .stMain{
  background-color: var(--bg_clr)
 }

 .stSidebar{
  background-color: var(--bg_clr_g2);
 }

  .stTextArea textarea, .stTextArea > textarea {
  min-height: 250px;
  font-size: 16px;
  background-color: var(--bg_clr_g2);
  border-color: var(--border_clr);
  resize: none;
 }

 .responseCard {
  background-color: var(--bg_clr);
  border-radius: 10px;
  padding: 1.5rem;
  margin: 1rem 0;
  border-left: 4px solid var(--primary_clr);
 }
 
 .stButton button, .stButton > button {
  background-color: var(--primary_clr);
  color: white;
  transition: all 0.3s;
  border: 1px solid var(--border_clr);
 }
 
 .stButton button:hover, .stButton button:focus:not(:active) {
  opacity: 0.9;
  transform: translateY(-1px);
  color: white;
  background-color: var(--bg_clr_g3);
  border: 1px solid var(--border_clr);
 }

 /*
 .stSlider [role="slider"]{
  background-color: var(--bg_clr_g3);
 }

 .stSlider [role="slider"] .st-an, .stSlider [role="slider"] .st-ap, .stSlider [role="slider"] .st-aq,
 .stSlider [role="slider"] .st-ao, .stSlider [role="slider"] .st-cu, .stSlider [role="slider"] .st-cv,
 .stSlider [role="slider"] .st-am, .stSlider [role="slider"] .st-cw, .stSlider [role="slider"] .st-cx{
  background: linear-gradient(to right, rgb(27, 66, 46) 0%, 
  rgb(27, 66, 46) 72.2222%,
  rgba(151, 166, 195, 0.25) 72.2222%,
  rgba(151, 166, 195, 0.25) 100%) !important;
 }*/

 
 .progress-container {
  margin-top: -10px;
  margin-bottom: 15px;
 }

 summary[class*="st-emotion-cache-"]:hover{
  font-weight:800;
  color: var(--font_clr);
 }
 
 .crisis-alert {
  background-color: #fff8f8;
  border-left: 5px solid #ff4b4b;
  padding: 1.5rem;
  margin: 1rem 0;
  border-radius: 0 8px 8px 0;
  box-shadow: 0 2px 8px rgba(255, 75, 75, 0.15);
  animation: pulse 2s infinite;
 }
 
 @keyframes pulse {
  0% { border-left-color: #ff4b4b; }
  50% { border-left-color: #ff9999; }
  100% { border-left-color: #ff4b4b; }
 }
 
 .crisis-alert strong {
  color: #d10000;
 }
</style>
""", unsafe_allow_html=True)

# 3. MODEL LOADING ===================================================
@st.cache_resource(show_spinner=False)
def load_model():
    return pipeline("text2text-generation", model="google/flan-t5-base")


# 4. PROMPT ENGINEERING ==============================================
def generate_prompt(user_input, counseling_style):
    """Generate context-aware prompts for different counseling approaches"""
    style_prompts = {
        "Cognitive Behavior Therapy": (
            "As a CBT therapist, suggest techniques to address: '{input}'. "
            "Focus on identifying cognitive distortions and suggest behavioral experiments. "
            "Provide 2-3 concrete interventions."
        ),
        "Psychodynamic": (
            "From a psychodynamic perspective, analyze: '{input}'. "
            "Consider unconscious patterns and childhood influences. "
            "Suggest exploratory questions to reveal underlying conflicts."
        ),
        "Humanistic": (
            "Using humanistic approach, respond to: '{input}'. "
            "Focus on unconditional positive regard and self-actualization. "
            "Provide empathetic reflections and growth-oriented suggestions."
        ),
        "Solution-Focused": (
            "Using solution-focused therapy, address: '{input}'. "
            "Identify exceptions to the problem and small achievable steps. "
            "Suggest 2-3 scaling questions or miracle questions."
        )
    }
    return style_prompts[counseling_style].format(input=user_input)

def get_references(approach):
    """Return evidence-based references with clinical guidelines"""
    references = {
        "Cognitive Behavior Therapy": {
            "text": "Beck, J. S. (2011). Cognitive Behavior Therapy: Basics and Beyond",
            "guide": "https://www.apa.org/pubs/books/cognitive-behavior-therapy"
        },
        "Psychodynamic": {
            "text": "McWilliams, N. (2020). Psychoanalytic Diagnosis",
            "guide": "https://www.guilford.com/books/Psychoanalytic-Diagnosis/McWilliams/9781462543694"
        },
        "Humanistic": {
            "text": "Rogers, C. (1951). Client-Centered Therapy",
            "guide": "https://www.nationalcounsellingsociety.org/about-therapy/types/humanistic"
        },
        "Solution-Focused": {
            "text": "De Shazer, S. (1988). Clues: Investigating Solutions in Brief Therapy",
            "guide": "https://www.solutionfocused.net/what-is-sfbt/"
        }
    }
    ref = references.get(approach, {
        "text": "Evidence-Based Practice in Psychology",
        "guide": "https://www.apa.org/practice/guidelines/evidence-based"
    })
    return f"{ref['text']} | [Clinical Guidelines]({ref['guide']})"

# 5. MAIN APPLICATION ================================================
def main():
 # Initialize session history
 if 'history' not in st.session_state:
     st.session_state.history = []
 
 st.title("🧠 Counselor Guidance Assistant")
 st.markdown("""
 *Professional support for mental health practitioners*  

 Enter a patient scenario below for evidence-based intervention suggestions.
 """)
 
 # Sidebar configuration
 with st.sidebar:
  st.title("Session Settings")
  counseling_style = st.selectbox(
      "Therapeutic Approach",
      ["Cognitive Behavior Therapy", "Psychodynamic", "Humanistic", "Solution-Focused"],
      index=0
  )
  creativity = st.slider("Response Creativity", 0.1, 1.0, 0.7)
  st.markdown("---")
  
  st.caption("""
  **Best Practices:**
  1. Describe specific behaviors/symptoms
  2. Include relevant history
  3. Note attempted interventions
  """)

  # Session history viewer
  if st.session_state.history:
   with st.expander("📚 Session History (Last 5)"):
    for i, session in enumerate(st.session_state.history[-5:][::-1]):
        st.markdown(f"""
        **Session {len(st.session_state.history)-i}** ({session['timestamp']})
        - Approach: {session['approach']}
        - Case: {session['case']}
        """)
        if st.button(f"View Details #{len(st.session_state.history)-i}", key=f"view_{i}"):
            st.session_state.current_session = session
     
 # Example cases for quick testing
 with st.expander("💡 Quick Start : Explore most commons challenges!! "):
  examples = {
    "Depression": "45yo male with treatment-resistant depression, expresses hopelessness about ever improving",
    "Anxiety": "College student experiencing panic attacks before exams despite knowing the material well",
    "Relationship": "Couple stuck in pursue-withdraw pattern, escalating arguments about household responsibilities"
  }
  cols = st.columns(3)
  for i, (label, example) in enumerate(examples.items()):
   with cols[i]:
    if st.button(label):
     case_description = example    
 
 # Main input area
 case_description = st.text_area(
  "Type your challenge below:",
  placeholder="My 28-year-old patient with social anxiety avoids all group situations despite previous exposure work...",
  height=250
 )
 
 # Crisis keywords detection
 CRISIS_KEYWORDS = ['suicide', 'self-harm', 'homicide', 'abuse', 'abused', 'kill myself', 'kill', 
          'want to die', 'end my life', 'hurt myself', 'hurt someone','suicidal']
 
 # Response generation
 if st.button("Analyze && Suggest", type="primary"):
  if not case_description.strip():
   st.warning("Please describe the clinical situation")
  else:
   # Crisis detection
   if any(keyword in case_description.lower() for keyword in CRISIS_KEYWORDS):
    st.markdown("""
    <div class="crisis-alert">
        <div style="font-size: 1.3rem;">⚠️ <strong>CRISIS ALERT</strong> - Immediate Action Required</div>
        <br>
        <div><strong>Clinical Protocols:</strong></div>
        <ol>
            <li>Assess immediate safety risk using direct questioning</li>
            <li>Implement safety planning if risk is present</li>
            <li>Do not leave patient alone if active suicidal/homicidal ideation exists</li>
        </ol>
        <div><strong>Emergency Resources:</strong></div>
        <ul>
            <li>🇺🇸 <strong>988 Suicide & Crisis Lifeline</strong> (24/7)</li>
            <li>📱 <strong>Crisis Text Line</strong>: Text HOME to 741741</li>
            <li>🌎 <strong>International Association for Suicide Prevention</strong>: <a href="https://www.iasp.info/resources/Crisis_Centres/">Find Local Help</a></li>
        </ul>
    </div>
    """, unsafe_allow_html=True)
    
    with st.expander("📋 Clinician Guidance (Click for Protocol Details)"):
     st.markdown("""
     **Standard Crisis Response Protocol:**
     1. **Direct Assessment**  
        "Are you having thoughts of ending your life?"  
        "Do you have a plan?"  
        "Have you ever attempted before?"
     
     2. **Safety Planning**  
        - Remove access to means  
        - Identify support contacts  
        - Create step-by-step coping strategies
     
     3. **Documentation**  
        - Risk assessment findings  
        - Actions taken  
        - Follow-up plan
     """)
    
    # Log crisis event
    st.session_state.history.append({
        'timestamp': datetime.now().strftime("%Y-%m-%d %H:%M"),
        'approach': "CRISIS INTERVENTION",
        'case': "CRISIS DETECTED - " + case_description[:50] + "...",
        'recommendations': "Session halted - emergency protocols activated"
    })
    st.stop()
   
   with st.spinner("Generating evidence-based suggestions..."):
    try:
        llm = load_model()
        prompt = generate_prompt(case_description, counseling_style)
        
        # Simulate processing steps for better UX
        progress_bar = st.progress(0)
        for percent in range(0, 101, 20):
            time.sleep(0.1)
            progress_bar.progress(percent)
        
        response = llm(
            prompt,
            max_length=500,
            do_sample=True,
            temperature=creativity
        )[0]['generated_text']
        
        # Store session in history
        st.session_state.history.append({
            'timestamp': datetime.now().strftime("%Y-%m-%d %H:%M"),
            'approach': counseling_style,
            'case': case_description[:100] + "..." if len(case_description) > 100 else case_description,
            'recommendations': response
        })
        
        # Display formatted response
        st.markdown("## Clinical Recommendations")
        with st.container():
            st.markdown(f'<div class="responseCard">{response}</div>', 
                       unsafe_allow_html=True)
            
            # Add references
            st.markdown("---")
            st.caption(f"**Reference:** {get_references(counseling_style)}")
            
            # Response tools
            st.download_button(
                "Save Recommendations",
                data=f"Approach: {counseling_style}\n\n{response}",
                file_name=f"clinical_suggestions_{counseling_style}.txt"
            )
        
    except Exception as e:
        st.error(f"Error generating suggestions: {str(e)}")

if __name__ == "__main__":
    main()