src/streamlit_app.py

#!/usr/bin/env python3
"""
Simplified Hugging Face Spaces compatible Multi-Agent System Dashboard
"""

import os
import sys
import tempfile
import sqlite3
from pathlib import Path
import warnings
from datetime import datetime, timedelta
import random

# Suppress warnings
warnings.filterwarnings('ignore')

# Set environment variables for Hugging Face Spaces
os.environ['STREAMLIT_SERVER_HEADLESS'] = 'true'
os.environ['STREAMLIT_SERVER_PORT'] = '7860'
os.environ['STREAMLIT_BROWSER_GATHER_USAGE_STATS'] = 'false'

# Import streamlit first and set page config
import streamlit as st

st.set_page_config(
    page_title="🤖 Multi-Agent System Dashboard",
    page_icon="🤖",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Import other required modules
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
import json
import numpy as np
from typing import Dict, List, Any

class SimpleDashboard:
    def __init__(self):
        # Use temp directory for database
        temp_dir = tempfile.gettempdir()
        self.db_path = os.path.join(temp_dir, "evaluation_logs.db")
        try:
            self.setup_demo_data()
        except Exception as e:
            st.error(f"Setup error: {str(e)}")
            self.create_fallback_data()
    
    def create_fallback_data(self):
        """Create fallback data if database fails"""
        st.warning("Using fallback demo data")
        
        # Create sample data directly
        agents = ["Diet Agent", "Support Agent", "Queries Agent"]
        data = []
        
        for i in range(50):
            base_score = random.uniform(7.0, 9.5)
            accuracy = random.uniform(7.0, 9.5)
            data.append({
                'id': i,
                'session_id': f"session_{random.randint(1000, 9999)}",
                'agent_name': random.choice(agents),
                'query': f"Sample query {i}",
                'response': f"Sample response {i} with detailed information and comprehensive guidance...",
                'overall_score': base_score,
                'relevance_score': random.uniform(7.0, 9.5),
                'accuracy_score': accuracy,
                'completeness_score': random.uniform(7.0, 9.5),
                'coherence_score': random.uniform(7.0, 9.5),
                'hallucination_score': max(0, min(10, 10 - accuracy + random.uniform(-1.0, 1.0))),
                'guardrails_passed': True,
                'safety_score': random.uniform(8.0, 10.0),
                'execution_time_ms': random.uniform(500, 2000),
                'input_tokens': random.randint(20, 100),
                'output_tokens': random.randint(100, 500),
                'total_tokens': random.randint(120, 600),
                'cost_usd': random.uniform(0.001, 0.02),
                'llm_provider': random.choice(["azure", "openai", "anthropic"]),
                'model_name': 'gpt-4o',
                'timestamp': datetime.now() - timedelta(days=random.randint(0, 30))
            })
        
        self.fallback_df = pd.DataFrame(data)
        self.use_fallback = True
    
    def setup_demo_data(self):
        """Setup demo database"""
        self.use_fallback = False
        
        if not os.path.exists(self.db_path):
            self.create_demo_database()
    
    def create_demo_database(self):
        """Create demo database"""
        conn = sqlite3.connect(self.db_path)
        cursor = conn.cursor()
        
        # Create table
        cursor.execute('''
        CREATE TABLE IF NOT EXISTS evaluation_logs (
            id INTEGER PRIMARY KEY AUTOINCREMENT,
            session_id TEXT NOT NULL,
            agent_name TEXT NOT NULL,
            query TEXT NOT NULL,
            response TEXT,
            overall_score REAL,
            relevance_score REAL,
            accuracy_score REAL,
            completeness_score REAL,
            coherence_score REAL,
            hallucination_score REAL,
            guardrails_passed BOOLEAN,
            safety_score REAL,
            execution_time_ms REAL,
            input_tokens INTEGER,
            output_tokens INTEGER,
            total_tokens INTEGER,
            cost_usd REAL,
            llm_provider TEXT,
            model_name TEXT,
            timestamp DATETIME DEFAULT CURRENT_TIMESTAMP
        )
        ''')
        
        # Insert demo data
        agents = ["Diet Agent", "Support Agent", "Queries Agent"]
        
        sample_queries = {
            "Diet Agent": [
                "What's a healthy meal plan for weight loss?",
                "Can you suggest low-carb breakfast options?",
                "What are the benefits of intermittent fasting?",
                "How much protein should I eat daily?",
                "What foods are good for heart health?"
            ],
            "Support Agent": [
                "I'm having trouble sleeping, can you help?",
                "How do I manage work stress?",
                "I feel overwhelmed with my tasks",
                "Can you help me organize my schedule?",
                "How to improve my productivity?"
            ],
            "Queries Agent": [
                "What are the latest developments in AI?",
                "How does blockchain technology work?",
                "What is quantum computing?",
                "Explain machine learning algorithms",
                "What are the benefits of cloud computing?"
            ]
        }
        
        for i in range(100):
            session_id = f"session_{random.randint(1000, 9999)}"
            agent = random.choice(agents)
            query = random.choice(sample_queries[agent])
            
            # Generate comprehensive response
            response_templates = {
                "Diet Agent": [
                    "Thank you for your question about nutrition and dietary guidance. I'd be happy to help you develop a healthier relationship with food and create sustainable eating habits.",
                    "I understand you're looking for dietary advice, and I'm here to provide evidence-based nutritional guidance tailored to your specific needs and goals."
                ],
                "Support Agent": [
                    "I appreciate you reaching out for support. It takes courage to ask for help, and I'm here to provide you with practical strategies and emotional guidance.",
                    "Thank you for sharing your concerns with me. I understand this can be challenging, and I want to help you work through this step by step with compassion and understanding."
                ],
                "Queries Agent": [
                    "Excellent question! This is a fascinating topic that involves cutting-edge technology and has significant implications for our future. Let me provide you with a comprehensive overview.",
                    "Thank you for this thought-provoking question. This subject encompasses multiple disciplines and recent innovations. I'll break this down into key concepts and practical applications."
                ]
            }
            
            base_response = random.choice(response_templates[agent])
            
            # Add detailed information
            if agent == "Diet Agent":
                details = "**Key Nutritional Recommendations:**\n\n1. **Whole Foods Focus**: Prioritize unprocessed foods like fresh fruits, vegetables, whole grains, lean proteins, and healthy fats.\n\n2. **Portion Control**: Use the plate method - fill half your plate with non-starchy vegetables, one quarter with lean protein, and one quarter with complex carbohydrates.\n\n3. **Hydration**: Aim for 8-10 glasses of water daily to support metabolism and overall health."
            elif agent == "Support Agent":
                details = "**Comprehensive Support Strategy:**\n\n**Immediate Coping Techniques:**\n1. **Deep Breathing**: Practice the 4-7-8 technique - inhale for 4 counts, hold for 7, exhale for 8.\n\n2. **Grounding Exercises**: Use the 5-4-3-2-1 method - identify 5 things you can see, 4 you can touch, 3 you can hear, 2 you can smell, and 1 you can taste.\n\n**Long-term Strategies:**\n- Establish a consistent daily routine\n- Practice mindfulness meditation for 10-15 minutes daily"
            else:  # Queries Agent
                details = "**Technical Deep Dive:**\n\n**Fundamental Concepts:**\nThis technology represents a convergence of multiple disciplines including computer science, mathematics, engineering, and domain-specific expertise.\n\n**Current Implementation:**\n1. **Healthcare**: AI-powered diagnostic tools and personalized treatment plans\n2. **Finance**: Algorithmic trading and fraud detection\n3. **Transportation**: Autonomous vehicles and traffic optimization"
            
            response = f"{base_response}\n\n{details}"
            
            # Generate realistic scores
            base_score = random.uniform(7.0, 9.5)
            relevance_score = max(0, min(10, base_score + random.uniform(-0.3, 0.3)))
            accuracy_score = max(0, min(10, base_score + random.uniform(-0.4, 0.2)))
            completeness_score = max(0, min(10, base_score + random.uniform(-0.5, 0.3)))
            coherence_score = max(0, min(10, base_score + random.uniform(-0.2, 0.4)))
            hallucination_score = max(0, min(10, 10 - accuracy_score + random.uniform(-1.0, 1.0)))
            
            # Generate token consumption
            response_length = len(response)
            input_tokens = int(len(query.split()) * 1.3)
            output_tokens = int(response_length / 4)
            total_tokens = input_tokens + output_tokens
            
            # Calculate cost
            llm_provider = random.choice(["azure", "openai", "anthropic"])
            cost_per_1k = {"azure": 0.03, "openai": 0.03, "anthropic": 0.025}
            cost_usd = (total_tokens / 1000) * cost_per_1k[llm_provider]
            
            timestamp = datetime.now() - timedelta(days=random.randint(0, 30))
            
            cursor.execute('''
            INSERT INTO evaluation_logs (
                session_id, agent_name, query, response, overall_score,
                relevance_score, accuracy_score, completeness_score, coherence_score,
                hallucination_score, guardrails_passed, safety_score, execution_time_ms,
                input_tokens, output_tokens, total_tokens, cost_usd, llm_provider, model_name, timestamp
            ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
            ''', (
                session_id, agent, query, response, base_score,
                relevance_score, accuracy_score, completeness_score, coherence_score,
                hallucination_score, random.choice([True, True, True, False]),  # 75% pass rate
                random.uniform(8.0, 10.0), random.uniform(500, 2000),
                input_tokens, output_tokens, total_tokens, round(cost_usd, 4),
                llm_provider, "gpt-4o", timestamp.isoformat()
            ))
        
        conn.commit()
        conn.close()
    
    def load_data(self):
        """Load data"""
        if self.use_fallback:
            return self.fallback_df
        
        try:
            conn = sqlite3.connect(self.db_path)
            df = pd.read_sql_query("SELECT * FROM evaluation_logs ORDER BY timestamp DESC", conn)
            conn.close()
            
            if not df.empty:
                df['timestamp'] = pd.to_datetime(df['timestamp'])
            
            return df
        except Exception as e:
            st.error(f"Data loading error: {str(e)}")
            return pd.DataFrame()
    
    def show_overview(self, df):
        """Show overview tab"""
        st.header("📈 Executive Summary")
        
        if df.empty:
            st.warning("No data available")
            return
        
        # Key metrics
        col1, col2, col3, col4 = st.columns(4)
        
        with col1:
            st.metric("Total Evaluations", len(df))
        
        with col2:
            avg_score = df['overall_score'].mean()
            st.metric("Average Score", f"{avg_score:.2f}/10")
        
        with col3:
            safety_rate = (df['guardrails_passed'].sum() / len(df)) * 100
            st.metric("Safety Rate", f"{safety_rate:.1f}%")
        
        with col4:
            avg_time = df['execution_time_ms'].mean() / 1000
            st.metric("Avg Response Time", f"{avg_time:.2f}s")
        
        # Charts
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("📊 Performance by Agent")
            agent_scores = df.groupby('agent_name')['overall_score'].mean().reset_index()
            fig = px.bar(
                agent_scores, 
                x='agent_name', 
                y='overall_score',
                title="Average Score by Agent",
                color='overall_score',
                color_continuous_scale='viridis'
            )
            st.plotly_chart(fig, use_container_width=True)
        
        with col2:
            st.subheader("📈 Score Distribution")
            fig = px.histogram(
                df, 
                x='overall_score', 
                nbins=20,
                title="Score Distribution",
                color_discrete_sequence=['#1f77b4']
            )
            st.plotly_chart(fig, use_container_width=True)
    
    def show_agent_performance(self, df):
        """Show agent performance tab"""
        st.header("🤖 Agent Performance Analysis")
        
        if df.empty:
            st.warning("No data available")
            return
        
        # Agent selector
        agents = df['agent_name'].unique()
        selected_agent = st.selectbox("Select Agent", ["All Agents"] + list(agents))
        
        # Filter data
        if selected_agent != "All Agents":
            filtered_df = df[df['agent_name'] == selected_agent]
        else:
            filtered_df = df
        
        # Performance metrics
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("🎯 Score Breakdown")
            score_cols = ['relevance_score', 'accuracy_score', 'completeness_score', 'coherence_score']
            available_scores = [col for col in score_cols if col in filtered_df.columns]
            
            if available_scores:
                avg_scores = filtered_df[available_scores].mean()
                fig = px.bar(
                    x=avg_scores.index,
                    y=avg_scores.values,
                    title=f"Average Scores - {selected_agent}",
                    labels={'x': 'Metric', 'y': 'Score'}
                )
                st.plotly_chart(fig, use_container_width=True)
        
        with col2:
            st.subheader("⏱️ Response Time Analysis")
            fig = px.box(
                filtered_df,
                x='agent_name',
                y='execution_time_ms',
                title="Response Time Distribution"
            )
            st.plotly_chart(fig, use_container_width=True)
        
        # Recent evaluations table
        st.subheader("📋 Recent Evaluations")
        display_cols = ['agent_name', 'query', 'overall_score', 'execution_time_ms', 'timestamp']
        available_cols = [col for col in display_cols if col in filtered_df.columns]
        
        if available_cols:
            recent_data = filtered_df[available_cols].head(20)
            st.dataframe(recent_data, use_container_width=True)
    
    def show_response_analysis(self, df):
        """Show response analysis tab"""
        st.header("📝 Response Analysis & Tracing")
        
        if df.empty:
            st.warning("No data available")
            return
        
        # Response metrics
        col1, col2, col3 = st.columns(3)
        
        with col1:
            if 'response' in df.columns:
                avg_length = df['response'].str.len().mean()
                st.metric("Avg Response Length", f"{avg_length:.0f} chars")
            else:
                st.metric("Avg Response Length", "N/A")
        
        with col2:
            if 'response' in df.columns:
                avg_words = df['response'].str.split().str.len().mean()
                st.metric("Avg Word Count", f"{avg_words:.0f} words")
            else:
                st.metric("Avg Word Count", "N/A")
        
        with col3:
            response_rate = (df['response'].notna().sum() / len(df)) * 100
            st.metric("Response Rate", f"{response_rate:.1f}%")
        
        # Search functionality
        st.subheader("🔍 Search Responses")
        search_term = st.text_input("Search in responses:", placeholder="Enter keywords...")
        
        if search_term and 'response' in df.columns:
            mask = df['response'].str.contains(search_term, case=False, na=False)
            search_results = df[mask]
        else:
            search_results = df.head(10)
        
        # Display results
        if not search_results.empty:
            st.write(f"**Showing {len(search_results)} results**")
            
            for idx, row in search_results.iterrows():
                with st.expander(f"🤖 {row['agent_name']} - Score: {row['overall_score']:.1f}"):
                    col1, col2 = st.columns([2, 1])
                    
                    with col1:
                        st.write("**Query:**")
                        st.write(row['query'])
                        
                        if 'response' in row and pd.notna(row['response']):
                            st.write("**Response:**")
                            st.write(row['response'])
                    
                    with col2:
                        st.write("**Metrics:**")
                        st.write(f"Overall Score: {row['overall_score']:.1f}/10")
                        if 'execution_time_ms' in row:
                            st.write(f"Response Time: {row['execution_time_ms']:.0f}ms")
                        if 'timestamp' in row:
                            st.write(f"Timestamp: {row['timestamp']}")
    
    def show_workflow_visualization(self, df):
        """Show workflow visualization tab"""
        st.header("🔄 Workflow Visualization")
        
        if df.empty:
            st.warning("No data available for workflow visualization.")
            return
        
        # Session selection
        sessions = df['session_id'].unique()
        selected_session = st.selectbox("Select Session", sessions, key="workflow_session")
        
        # Filter data for selected session
        session_data = df[df['session_id'] == selected_session]
        
        if session_data.empty:
            st.warning("No data found for selected session.")
            return
        
        # Session metrics overview
        st.subheader("📈 Session Metrics Overview")
        
        col1, col2, col3, col4 = st.columns(4)
        
        with col1:
            avg_score = session_data['overall_score'].mean()
            st.metric("Avg Overall Score", f"{avg_score:.2f}/10")
        
        with col2:
            avg_latency = session_data['execution_time_ms'].mean()
            st.metric("Avg Response Time", f"{avg_latency:.0f}ms")
        
        with col3:
            if 'hallucination_score' in session_data.columns:
                avg_hallucination = session_data['hallucination_score'].mean()
                st.metric("Avg Hallucination", f"{avg_hallucination:.2f}/10")
            else:
                st.metric("Avg Hallucination", "N/A")
        
        with col4:
            if 'total_tokens' in session_data.columns:
                total_tokens = session_data['total_tokens'].sum()
                total_cost = session_data['cost_usd'].sum() if 'cost_usd' in session_data.columns else 0
                st.metric("Total Cost", f"${total_cost:.4f}", f"{total_tokens:,} tokens")
            else:
                st.metric("Total Cost", "N/A")
        
        # Workflow steps
        st.subheader("🔍 Workflow Steps")
        
        for idx, (_, row) in enumerate(session_data.iterrows()):
            with st.expander(f"Step {idx + 1}: {row['agent_name']} - Score: {row['overall_score']:.2f}/10"):
                
                col1, col2 = st.columns([1, 1])
                
                with col1:
                    st.markdown("**Query:**")
                    st.write(row['query'])
                    
                    # Performance metrics chart
                    st.markdown("**Performance Metrics:**")
                    metrics_data = {
                        'Overall': row['overall_score'],
                        'Relevance': row.get('relevance_score', 0),
                        'Accuracy': row.get('accuracy_score', 0),
                        'Completeness': row.get('completeness_score', 0),
                        'Coherence': row.get('coherence_score', 0)
                    }
                    
                    if 'hallucination_score' in row:
                        metrics_data['Hallucination'] = row['hallucination_score']
                    
                    fig = px.bar(
                        x=list(metrics_data.keys()), 
                        y=list(metrics_data.values()),
                        title="Score Breakdown",
                        labels={'x': 'Metric', 'y': 'Score (0-10)'}
                    )
                    fig.update_layout(height=300, showlegend=False)
                    st.plotly_chart(fig, use_container_width=True)
                
                with col2:
                    st.markdown("**Response:**")
                    if pd.notna(row['response']):
                        st.write(row['response'])
                    else:
                        st.write("No response available")
                    
                    # Resource consumption
                    st.markdown("**Resource Consumption:**")
                    
                    if 'input_tokens' in row and pd.notna(row['input_tokens']):
                        token_col1, token_col2 = st.columns(2)
                        with token_col1:
                            st.metric("Input Tokens", f"{int(row['input_tokens']):,}")
                            st.metric("Output Tokens", f"{int(row.get('output_tokens', 0)):,}")
                        
                        with token_col2:
                            st.metric("Total Tokens", f"{int(row.get('total_tokens', 0)):,}")
                            st.metric("Cost", f"${row.get('cost_usd', 0):.4f}")
                    
                    # Execution details
                    st.markdown("**Execution Details:**")
                    st.write(f"⏱️ **Execution Time:** {row['execution_time_ms']:.0f}ms")
                    if 'llm_provider' in row:
                        st.write(f"🤖 **LLM Provider:** {row['llm_provider']}")
                    if 'model_name' in row:
                        st.write(f"🧠 **Model:** {row['model_name']}")
                    st.write(f"🛡️ **Safety Passed:** {'✅' if row['guardrails_passed'] else '❌'}")
        
        # Session summary
        st.subheader("📋 Session Summary")
        
        summary_col1, summary_col2, summary_col3 = st.columns(3)
        
        with summary_col1:
            st.markdown("**Quality Metrics:**")
            st.write(f"• Average Overall Score: {session_data['overall_score'].mean():.2f}/10")
            best_step = session_data.loc[session_data['overall_score'].idxmax()]
            st.write(f"• Best Performing Step: {best_step['agent_name']}")
            st.write(f"• Consistency (Std Dev): {session_data['overall_score'].std():.2f}")
        
        with summary_col2:
            st.markdown("**Performance Metrics:**")
            st.write(f"• Total Execution Time: {session_data['execution_time_ms'].sum():.0f}ms")
            st.write(f"• Average Response Time: {session_data['execution_time_ms'].mean():.0f}ms")
            st.write(f"• Fastest Step: {session_data['execution_time_ms'].min():.0f}ms")
        
        with summary_col3:
            st.markdown("**Resource Usage:**")
            if 'total_tokens' in session_data.columns:
                st.write(f"• Total Tokens Used: {session_data['total_tokens'].sum():,}")
                if 'cost_usd' in session_data.columns:
                    st.write(f"• Total Cost: ${session_data['cost_usd'].sum():.4f}")
                    st.write(f"• Avg Cost per Query: ${session_data['cost_usd'].mean():.4f}")
            else:
                st.write("• Token data not available")
        
        # Export functionality
        st.subheader("📤 Export Workflow Data")
        
        if st.button("Export Session Data to CSV", key="export_workflow"):
            csv_data = session_data.to_csv(index=False)
            st.download_button(
                label="Download CSV",
                data=csv_data,
                file_name=f"workflow_session_{selected_session}.csv",
                mime="text/csv"
            )
    
    def run(self):
        """Run the dashboard"""
        st.title("🤖 Multi-Agent System Dashboard")
        st.markdown("---")
        
        st.info("🎉 **Welcome!** This dashboard showcases evaluation metrics for Diet, Support, and Queries agents.")
        
        # Load data
        df = self.load_data()
        
        # Create tabs
        tab1, tab2, tab3, tab4 = st.tabs([
            "📈 Overview", 
            "🤖 Agent Performance", 
            "📝 Response Analysis",
            "🔄 Workflow Visualization"
        ])
        
        with tab1:
            self.show_overview(df)
        
        with tab2:
            self.show_agent_performance(df)
        
        with tab3:
            self.show_response_analysis(df)
        
        with tab4:
            self.show_workflow_visualization(df)
        
        # Footer
        st.markdown("---")
        st.markdown("🚀 **Multi-Agent System Dashboard** | Built with Streamlit & Plotly")

# Run the dashboard
try:
    dashboard = SimpleDashboard()
    dashboard.run()
except Exception as e:
    st.error(f"Application Error: {str(e)}")
    st.info("Please refresh the page.")
    
    with st.expander("Debug Information"):
        st.code(f"""
Error: {str(e)}
Type: {type(e).__name__}
Python: {sys.version}
Working Dir: {os.getcwd()}
Temp Dir: {tempfile.gettempdir()}
        """)