app.py

import base64
import io
import os
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt
from typing import Dict, List, Tuple, Any
import json
from litellm import completion
import logging

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# System prompts for different AI roles
CONVERSATION_PROMPT = """
You are an expert career advisor collecting information through natural conversation.
Focus on gathering these key points naturally without explicitly asking:
1. Current Role & Salary:
   - Current salary (must get specific number)
   - Job title/role
   - Company size/type
   - Industry/sector

2. Experience & Skills:
   - Years in current role
   - Total years working
   - Key skills and proficiencies
   - Management/leadership experience
   - Notable achievements

3. Education & Training:
   - Highest degree
   - Field of study
   - Certifications
   - Recent training/upskilling

4. Location & Work Setup:
   - Current location/market
   - Remote work status
   - Willingness to relocate
   - Preferred work arrangement

Conversation Guidelines:
- Be natural and conversational
- Show interest in their career journey
- Don't force structured responses
- Make salary discussions comfortable
- Build rapport through discussion
- Once you have sufficient information say:
  "I have collected enough information for analysis. Please click 'Generate Analysis' to see your career projections."
"""

EXTRACTION_PROMPT = """
Analyze the conversation and extract numerical scores based on salary growth potential.
Convert qualitative information into scores from 0 to 1.

Scoring Guidelines:

1. Industry Score (0-1):
   - 1.0: Cutting-edge tech (AI, quantum)
   - 0.9: High-growth tech (cloud, cyber)
   - 0.8: Established tech/finance
   - 0.7: Healthcare/biotech
   - 0.6: Traditional sectors
   - 0.5: Declining industries
   +0.1: Market leader company
   +0.1: High growth market

2. Experience Score (0-1):
   - 1.0: 15+ years with leadership
   - 0.9: 10-15 years senior
   - 0.8: 7-10 years mid-senior
   - 0.7: 4-6 years mid-level
   - 0.6: 2-3 years junior
   - 0.5: Entry level
   +0.1: Fast career progression
   +0.1: Significant achievements

3. Education Score (0-1):
   - 1.0: PhD from top school
   - 0.9: Masters from top school
   - 0.8: Bachelors from top school
   - 0.7: Advanced degree
   - 0.6: Bachelors degree
   - 0.5: Other education
   +0.1: Relevant certifications
   +0.1: Ongoing education

4. Skills Score (0-1):
   - 1.0: Rare, high-demand skills
   - 0.9: Advanced technical
   - 0.8: Strong tech + soft skills
   - 0.7: Solid technical skills
   - 0.6: Standard skills
   - 0.5: Basic skills
   +0.1: Multiple in-demand skills
   +0.1: Proven implementations

5. Location Score (0-1):
   - 1.0: Major tech hubs
   - 0.9: Secondary tech hubs
   - 0.8: Major cities
   - 0.7: Growing cities
   - 0.6: Regional cities
   - 0.5: Small markets
   +0.1: Remote flexibility
   +0.1: High growth market

Return only a JSON object with these exact fields:
{
    "industry_score": float,
    "experience_score": float,
    "education_score": float,
    "skills_score": float,
    "location_score": float,
    "current_salary": float
}

Make reasonable assumptions for any missing information based on context clues in the conversation.
"""

class CodeEnvironment:
    """Safe environment for executing visualization code"""
    
    def __init__(self):
        self.globals = {
            'np': np,
            'plt': plt
        }
        self.locals = {}
        
    def execute(self, code: str) -> Dict[str, Any]:
        """Execute code and capture outputs"""
        result = {'figures': [], 'error': None}
        
        try:
            # Execute code in safe environment
            exec(code, self.globals, self.locals)
            
            # Capture generated plots
            for i in plt.get_fignums():
                fig = plt.figure(i)
                buf = io.BytesIO()
                fig.savefig(buf, format='png', dpi=300)
                buf.seek(0)
                img_str = base64.b64encode(buf.read()).decode()
                result['figures'].append(f"data:image/png;base64,{img_str}")
                plt.close(fig)
                
        except Exception as e:
            result['error'] = str(e)
            
        return result

class MonteCarloSimulator:
    """Handles Monte Carlo simulation for salary projections"""
    
    def __init__(self):
        self.years = 5
        self.num_paths = 1000
        
    def run_simulation(self, profile: Dict[str, float]) -> np.ndarray:
        """Run Monte Carlo simulation for salary projections"""
        # Initialize paths array
        paths = np.zeros((self.num_paths, self.years + 1))
        paths[:, 0] = float(profile['current_salary'])
        
        # Calculate growth parameters from profile
        params = self._calculate_parameters(profile)
        
        # Generate paths
        for path in range(self.num_paths):
            salary = paths[path, 0]
            for year in range(1, self.years + 1):
                # Calculate growth with randomness
                growth = self._calculate_growth(params)
                # Update salary
                salary *= (1 + growth)
                paths[path, year] = salary
                
        return paths
    
    def _calculate_parameters(self, profile: Dict[str, float]) -> Dict[str, float]:
        """Calculate simulation parameters from profile"""
        return {
            'base_growth': 0.02 + (profile['industry_score'] * 0.04),
            'skill_premium': 0.01 + (profile['skills_score'] * 0.02),
            'experience_premium': 0.01 + (profile['experience_score'] * 0.02),
            'education_premium': 0.005 + (profile['education_score'] * 0.015),
            'location_premium': 0.01 + (profile['location_score'] * 0.02),
            'volatility': 0.05 + (profile['industry_score'] * 0.05),
            'disruption_chance': 0.1,
            'disruption_impact': 0.2
        }
    
    def _calculate_growth(self, params: Dict[str, float]) -> float:
        """Calculate annual growth rate with randomness"""
        # Base growth plus premiums
        growth = (params['base_growth'] + 
                 params['skill_premium'] + 
                 params['experience_premium'] + 
                 params['education_premium'] + 
                 params['location_premium'])
        
        # Add random volatility
        growth += np.random.normal(0, params['volatility'])
        
        # Possible disruption event
        if np.random.random() < params['disruption_chance']:
            disruption = params['disruption_impact'] * np.random.random()
            if np.random.random() < 0.7:  # 70% positive disruption
                growth += disruption
            else:
                growth -= disruption
        
        # Apply reasonable bounds
        return max(min(growth, 0.25), -0.1)  # -10% to +25%

class CareerAdvisor:
    """Main career advisor system"""
    
    def __init__(self, api_key: str):
        self.api_key = api_key
        self.chat_history = []
        self.code_env = CodeEnvironment()
        self.simulator = MonteCarloSimulator()
        
    def chat(self, message: str) -> str:
        """Handle conversation with user"""
        messages = [
            {"role": "system", "content": CONVERSATION_PROMPT},
            *self.chat_history,
            {"role": "user", "content": message}
        ]
        
        response = completion(
            model="gpt-4o-mini",
            messages=messages,
            api_key=self.api_key
        )
        
        self.chat_history.extend([
            {"role": "user", "content": message},
            {"role": "assistant", "content": response.choices[0].message.content}
        ])
        
        return response.choices[0].message.content
        
    def extract_profile(self) -> Dict[str, float]:
        """Extract numerical profile from conversation"""
        conversation = "\n".join([
            f"{msg['role']}: {msg['content']}"
            for msg in self.chat_history
        ])
        
        messages = [
            {"role": "system", "content": EXTRACTION_PROMPT},
            {"role": "user", "content": f"Extract profile from:\n{conversation}"}
        ]
        
        response = completion(
            model="gpt-4o-mini",
            messages=messages,
            api_key=self.api_key,
            response_format={"type": "json_object"}
        )
        
        return json.loads(response.choices[0].message.content)
        
    def generate_visualization(self, paths: np.ndarray) -> str:
        """Generate visualization of simulation results"""
        viz_code = """
        plt.style.use('dark_background')
        fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(12, 10), height_ratios=[2, 1])
        fig.tight_layout(pad=4)
        
        # Plot 1: Salary Projections
        years = list(range(paths.shape[1]))
        
        # Plot confidence intervals
        percentiles = [(5, 95), (10, 90), (25, 75)]
        alphas = [0.1, 0.2, 0.3]
        for (lower, upper), alpha in zip(percentiles, alphas):
            lower_bound = np.percentile(paths, lower, axis=0)
            upper_bound = np.percentile(paths, upper, axis=0)
            ax1.fill_between(years, lower_bound, upper_bound, alpha=alpha, color='blue')
        
        # Plot median line
        median = np.percentile(paths, 50, axis=0)
        ax1.plot(years, median, color='white', linewidth=2, label='Expected Path')
        
        # Customize plot
        ax1.set_title('Salary Growth Projections', pad=20)
        ax1.set_xlabel('Years')
        ax1.set_ylabel('Salary ($)')
        ax1.grid(True, alpha=0.2)
        ax1.legend()
        
        # Format axes
        ax1.yaxis.set_major_formatter(plt.FuncFormatter(lambda x, p: f'${x:,.0f}'))
        ax1.set_xticks(years)
        ax1.set_xticklabels(['Current'] + [f'Year {i+1}' for i in range(len(years)-1)])
        
        # Plot 2: Final Distribution
        ax2.hist(paths[:, -1], bins=50, color='blue', alpha=0.7)
        ax2.set_title('Final Salary Distribution', pad=20)
        ax2.set_xlabel('Salary ($)')
        ax2.set_ylabel('Frequency')
        ax2.grid(True, alpha=0.2)
        ax2.xaxis.set_major_formatter(plt.FuncFormatter(lambda x, p: f'${x:,.0f}'))
        """
        
        result = self.code_env.execute(viz_code)
        return result['figures'][0] if result['figures'] else None
        
    def generate_summary(self, profile: Dict[str, float], paths: np.ndarray) -> str:
        """Generate analysis summary"""
        final_salaries = paths[:, -1]
        initial_salary = paths[0, 0]
        cagr = (np.median(final_salaries) / initial_salary) ** (1/5) - 1
        
        return f"""
        Career Profile Analysis
        ======================

        Current Situation:
        • Salary: ${profile['current_salary']:,.2f}
        • Industry Position: {profile['industry_score']:.2f}/1.0
        • Experience Level: {profile['experience_score']:.2f}/1.0
        • Education Rating: {profile['education_score']:.2f}/1.0
        • Skills Assessment: {profile['skills_score']:.2f}/1.0
        • Location Impact: {profile['location_score']:.2f}/1.0

        5-Year Projection:
        • Conservative (25th percentile): ${np.percentile(final_salaries, 25):,.2f}
        • Most Likely (Median): ${np.percentile(final_salaries, 50):,.2f}
        • Optimistic (75th percentile): ${np.percentile(final_salaries, 75):,.2f}
        • Expected Annual Growth: {cagr*100:.1f}%

        Key Insights:
        • Your profile suggests {cagr*100:.1f}% annual growth potential
        • {profile['industry_score']:.2f} industry score indicates {'strong' if profile['industry_score'] > 0.7 else 'moderate' if profile['industry_score'] > 0.5 else 'challenging'} growth environment
        • Skills rating of {profile['skills_score']:.2f} suggests {'excellent' if profile['skills_score'] > 0.7 else 'good' if profile['skills_score'] > 0.5 else 'potential for'} career advancement
        • Location score {profile['location_score']:.2f} {'enhances' if profile['location_score'] > 0.7 else 'supports' if profile['location_score'] > 0.5 else 'may limit'} opportunities

        Based on {paths.shape[0]:,} simulated career paths
        """
        
    def generate_analysis(self) -> Tuple[str, str]:
        """Generate complete analysis with visualization"""
        try:
            # Extract profile from conversation
            profile = self.extract_profile()
            
            # Run Monte Carlo simulation
            paths = self.simulator.run_simulation(profile)
            
            # Generate visualization
            viz = self.generate_visualization(paths)
            
            # Generate summary
            summary = self.generate_summary(profile, paths)
            
            return summary, viz
            
        except Exception as e:
            return f"Error generating analysis: {str(e)}", None

def create_interface():
    """Create the Gradio interface"""
    advisor = None
    
    def init_advisor(api_key: str):
        nonlocal advisor
        if not api_key.strip().startswith("sk-"):
            return "Invalid API key format. Please check your key."
        advisor = CareerAdvisor(api_key)
        return "Advisor initialized! Let's discuss your career."
    
    def chat(message: str, history: List):
        if not advisor:
            return "Please enter your API key first.", history
        
        response = advisor.chat(message)
        history.append((message, response))
        return "", history
    
    def analyze():
        if not advisor:
            return "Please enter your API key first.", None
        
        summary, viz = advisor.generate_analysis()
        return summary, viz
    
    # Create interface
    with gr.Blocks() as demo:
        gr.Markdown("# AI Career Advisor with Monte Carlo Salary Projections")
        
        with gr.Row():
            api_key = gr.Textbox(
                label="OpenAI API Key",
                type="password",
                placeholder="Enter your API key"
            )
            init_btn = gr.Button("Initialize Advisor")
        
        status = gr.Textbox(label="Status")
        
        with gr.Row():
            with gr.Column(scale=1):
                chatbot = gr.Chatbot(height=400)
                msg = gr.Textbox(
                    label="Your message",
                    placeholder="Tell me about your career...",
                    lines=2
                )
                analyze_btn = gr.Button("Generate Analysis", variant="primary")
            
            with gr.Column(scale=1):
                analysis = gr.Textbox(
                    label="Analysis Report",
                    lines=20,
                    show_copy_button=True
                )
                plot = gr.Image(
                    label="Salary Projections",
                    show_download_button=True
                )
        
        # Wire up the interface
        init_btn.click(
            init_advisor,
            inputs=[api_key],
            outputs=[status]
        )
        
        msg.submit(
            chat,
            inputs=[msg, chatbot],
            outputs=[msg, chatbot]
        )
        
        analyze_btn.click(
            analyze,
            outputs=[analysis, plot]
        )
    
    return demo

if __name__ == "__main__":
    demo = create_interface()
    demo.launch()