## Ultimate Smart ATS System - June 2025 Edition
## Incorporates Latest AI Strategies with Deterministic Consistency

import streamlit as st
import os
import re
import tempfile
import sqlite3
import hashlib
import json
import time
import pickle
from datetime import datetime
from pathlib import Path
from functools import wraps
from PyPDF2 import PdfReader
import google.generativeai as genai
from dotenv import load_dotenv

# Optional imports with fallbacks
try:
    import pdfplumber
    HAS_PDFPLUMBER = True
except ImportError:
    HAS_PDFPLUMBER = False

try:
    import docx
    HAS_DOCX = True
except ImportError:
    HAS_DOCX = False

try:
    import tiktoken
    HAS_TIKTOKEN = True
except ImportError:
    HAS_TIKTOKEN = False

# Loading the .env keys
load_dotenv()

# Create cache directory
CACHE_DIR = Path("ats_cache")
CACHE_DIR.mkdir(exist_ok=True)

# MASTER UNIVERSAL SYSTEM PROMPT - Designed for Maximum Consistency & Global Applicability
UNIVERSAL_MASTER_PROMPT = """
You are the ULTIMATE ATS OPTIMIZATION ENGINE 3.0 - A state-of-the-art AI system designed to provide CONSISTENT, PRECISE, and GLOBALLY APPLICABLE resume analysis across ALL industries, roles, and experience levels.

**CORE OPERATING PRINCIPLES (June 2025 Standards):**

**CONSISTENCY PROTOCOL:**
- Always follow the EXACT same evaluation methodology for identical inputs
- Use standardized scoring matrices to ensure reproducible results
- Apply deterministic analysis patterns to eliminate variability
- Maintain consistent terminology and assessment criteria across all evaluations

**UNIVERSAL EVALUATION FRAMEWORK:**

**PHASE 1: TECHNICAL PARSING & KEYWORD OPTIMIZATION (25 Points)**
1. **ATS Compatibility Analysis:**
   - Format compliance score (reverse-chronological preferred)
   - Keyword density and natural integration assessment
   - Critical missing terms identification (high-impact keywords only)
   - Semantic relevance evaluation

2. **Section-by-Section Parsing:**
   - Contact Information: Completeness and professionalism
   - Professional Summary: Impact and relevance
   - Experience: Achievement quantification and progression
   - Skills: Technical and soft skills alignment
   - Education: Relevance and credential verification

**PHASE 2: COMPETENCY MATCHING ANALYSIS (35 Points)**
1. **Hard Skills Assessment (20 points):**
   - Technical competencies directly matching job requirements
   - Tools, software, and platform expertise
   - Industry-specific knowledge and certifications
   - Quantifiable achievements and metrics

2. **Soft Skills Evaluation (15 points):**
   - Leadership and communication indicators
   - Problem-solving and analytical thinking evidence
   - Adaptability and learning agility markers
   - Team collaboration and project management skills

**PHASE 3: EXPERIENCE RELEVANCE SCORING (25 Points)**
1. **Direct Experience Match (15 points):**
   - Same industry and role experience
   - Progressive responsibility growth
   - Relevant project complexity and scale

2. **Transferable Experience (10 points):**
   - Adjacent industry or functional experience
   - Cross-domain skills applicability
   - Leadership and management experience

**PHASE 4: CULTURAL & GROWTH POTENTIAL (15 Points)**
1. **Growth Trajectory Analysis:**
   - Career progression patterns
   - Continuous learning evidence
   - Innovation and initiative indicators

2. **Cultural Alignment Markers:**
   - Values demonstration through experience
   - Collaboration and team success stories
   - Adaptability to organizational change

**STANDARDIZED OUTPUT STRUCTURE:**
Always provide results in this EXACT format for consistency:

**🎯 OVERALL MATCH SCORE: [XX/100]**

**📊 DETAILED BREAKDOWN:**
- Technical Parsing: [XX/25]
- Competency Matching: [XX/35] 
- Experience Relevance: [XX/25]
- Growth Potential: [XX/15]

**✅ TOP 5 STRENGTHS:**
1. [Specific strength with evidence]
2. [Specific strength with evidence]
3. [Specific strength with evidence]
4. [Specific strength with evidence]
5. [Specific strength with evidence]

**🔧 TOP 3 IMPROVEMENT AREAS:**
1. [Specific area with actionable suggestion]
2. [Specific area with actionable suggestion]
3. [Specific area with actionable suggestion]

**🔑 CRITICAL MISSING KEYWORDS:**
- [High-impact keyword 1]
- [High-impact keyword 2]
- [High-impact keyword 3]

**💡 STRATEGIC RECOMMENDATIONS:**
[3-4 specific, actionable recommendations]

**🏆 FINAL VERDICT:**
[EXCEPTIONAL 90-100 | STRONG 75-89 | GOOD 60-74 | DEVELOPING 45-59 | NEEDS WORK <45]

**CONSISTENCY GUARANTEES:**
- Same resume + same job description = identical analysis (±2 points variation max)
- Standardized language and terminology across all evaluations
- Reproducible scoring methodology regardless of domain
- Time-consistent results (same analysis today and tomorrow)
"""

# Specialized prompts that extend the master prompt for specific use cases
SPECIALIZED_PROMPTS = {
    "evaluate_resume": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: COMPREHENSIVE RESUME EVALUATION**
Apply the Universal Evaluation Framework above to provide a complete assessment.
Focus on overall candidacy evaluation with balanced perspective on strengths and development areas.
Maintain professional tone suitable for HR professionals and hiring managers.
""",

    "improve_skills": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: SKILL ENHANCEMENT STRATEGY**
After completing the standard evaluation, provide additional guidance:
**📈 SKILL DEVELOPMENT ROADMAP:**
- **Immediate Actions (0-3 months):** Quick wins and foundational improvements
- **Short-term Goals (3-12 months):** Structured learning and certification paths  
- **Long-term Vision (1-3 years):** Strategic career advancement opportunities
**🎓 LEARNING RESOURCES:**
- Recommended courses, certifications, and training programs
- Industry conferences and networking opportunities
- Practical projects and portfolio development suggestions
Focus on actionable, measurable improvement strategies with clear timelines.
""",

    "missing_keywords": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: ATS KEYWORD OPTIMIZATION**
After completing the standard evaluation, provide enhanced keyword analysis:
**🔍 ADVANCED KEYWORD ANALYSIS:**
- **CRITICAL MISSING (High Impact):** Essential terms significantly affecting ATS ranking
- **IMPORTANT ADDITIONS (Medium Impact):** Valuable terms improving visibility
- **OPTIMIZATION OPPORTUNITIES (Low Impact):** Supplementary terms for comprehensive coverage
**📝 INTEGRATION STRATEGY:**
- Specific resume sections for keyword placement
- Natural integration techniques avoiding keyword stuffing
- Industry-appropriate phrasing and terminology
**🤖 ATS COMPATIBILITY SCORE:** [Detailed breakdown of parsing efficiency]
""",

    "percentage_match": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: PRECISE MATCHING ANALYSIS**
Provide the standard evaluation with enhanced quantitative focus:
**📊 DETAILED SCORING BREAKDOWN:**
Present exact point allocation for each category with clear justification.
Include competitive benchmarking and market positioning analysis.
Provide specific improvement strategies for 10-15% score increase.
**🎯 MATCH PERCENTAGE: [XX%]**
Tier Classification with detailed rationale and next steps.
""",

    "answer_query": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: EXPERT CONSULTATION**
Apply domain expertise to answer the specific query while considering:
- Resume content and job description context
- Industry best practices and current market trends
- Practical, actionable guidance
- Evidence-based recommendations
Provide thorough, well-researched responses with specific examples and multiple solution approaches when applicable.
""",

    "executive_assessment": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: EXECUTIVE-LEVEL EVALUATION**
Apply enhanced criteria for senior leadership positions:
**👔 EXECUTIVE COMPETENCY FRAMEWORK:**
- Strategic thinking and vision development
- Change management and transformation leadership
- Financial acumen and business impact
- Board readiness and governance experience
**📈 LEADERSHIP IMPACT ANALYSIS:**
- Quantifiable business results and achievements
- Market expansion and competitive positioning
- Organizational culture and talent development
- Crisis leadership and resilience
Provide insights suitable for C-suite and board-level discussions.
""",

    "career_transition": f"""
{UNIVERSAL_MASTER_PROMPT}
**SPECIFIC TASK: CAREER PIVOT ANALYSIS**
Evaluate career change feasibility with:
**🔄 TRANSITION ASSESSMENT:**
- Transferable skills mapping across industries
- Market positioning strategy for career change
- Risk mitigation and success probability analysis
- Timeline and milestone planning
**🎯 TRANSITION ROADMAP:**
- Phase-wise transition strategy
- Skill development priorities
- Network building and industry immersion plan
Provide strategic guidance maximizing transition success while minimizing career risks.
"""
}

# Enhanced configuration for consistency
GENERATION_CONFIG = {
    "temperature": 0.15,  # Low temperature for maximum consistency
    "top_p": 0.8,
    "top_k": 40,
    "max_output_tokens": 4096,
    "stop_sequences": [],
}

# Model options optimized for consistency and performance
MODEL_FALLBACK_CHAIN = [
    "gemini-2.5-flash",                      # 🥇 PRIMARY - Latest Gemini 2.5
    "gemini-2.5-flash-lite-preview-06-17"   # 🥈 FALLBACK - Lite version
]

# Rate limiting decorator
def rate_limit(min_interval=2):
    def decorator(func):
        last_called = [0]
        @wraps(func)
        def wrapper(*args, **kwargs):
            elapsed = time.time() - last_called[0]
            left_to_wait = min_interval - elapsed
            if left_to_wait > 0:
                time.sleep(left_to_wait)
            result = func(*args, **kwargs)
            last_called[0] = time.time()
            return result
        return wrapper
    return decorator

# Cache management functions
def init_cache():
    """Initialize SQLite cache for consistency"""
    conn = sqlite3.connect(CACHE_DIR / "ats_cache.db")
    cursor = conn.cursor()
    cursor.execute("""
        CREATE TABLE IF NOT EXISTS analysis_cache (
            hash_key TEXT PRIMARY KEY,
            response TEXT,
            timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
            model_used TEXT
        )
    """)
    conn.commit()
    conn.close()

def get_cached_response(consistency_hash, model_id):
    """Get cached response if available"""
    try:
        conn = sqlite3.connect(CACHE_DIR / "ats_cache.db")
        cursor = conn.cursor()
        cursor.execute(
            "SELECT response FROM analysis_cache WHERE hash_key = ? AND model_used = ?",
            (consistency_hash, model_id)
        )
        result = cursor.fetchone()
        conn.close()
        return result[0] if result else None
    except:
        return None

def cache_response(consistency_hash, response, model_id):
    """Cache the response for future use"""
    try:
        conn = sqlite3.connect(CACHE_DIR / "ats_cache.db")
        cursor = conn.cursor()
        cursor.execute(
            "INSERT OR REPLACE INTO analysis_cache (hash_key, response, model_used) VALUES (?, ?, ?)",
            (consistency_hash, response, model_id)
        )
        conn.commit()
        conn.close()
    except Exception as e:
        st.warning(f"Cache save failed: {e}")

# Token estimation and content optimization
def estimate_tokens(text, model="gpt-3.5-turbo"):
    """Estimate token count for text"""
    if HAS_TIKTOKEN:
        try:
            encoding = tiktoken.encoding_for_model(model)
            return len(encoding.encode(text))
        except:
            pass
    # Fallback estimation: roughly 4 characters per token
    return len(text) // 4

def optimize_content_length(resume_text, job_description, max_resume_tokens=2000, max_job_tokens=1500):
    """Optimize content length to stay within token limits"""
    
    # Prioritize key sections in resume
    resume_sections = {
        'experience': [],
        'skills': [],
        'education': [],
        'summary': []
    }
    
    # Simple section detection
    lines = resume_text.split('\n')
    current_section = 'summary'
    
    for line in lines:
        line_lower = line.lower().strip()
        if any(keyword in line_lower for keyword in ['experience', 'work', 'employment']):
            current_section = 'experience'
        elif any(keyword in line_lower for keyword in ['skills', 'technical', 'competencies']):
            current_section = 'skills'
        elif any(keyword in line_lower for keyword in ['education', 'academic', 'degree']):
            current_section = 'education'
        
        if line.strip():
            resume_sections[current_section].append(line)
    
    # Build optimized resume content
    optimized_resume = []
    
    # Add summary (first 300 chars)
    if resume_sections['summary']:
        summary_text = '\n'.join(resume_sections['summary'][:5])
        optimized_resume.append(f"PROFESSIONAL SUMMARY:\n{summary_text[:300]}")
    
    # Add experience (prioritize recent)
    if resume_sections['experience']:
        exp_text = '\n'.join(resume_sections['experience'][:15])
        optimized_resume.append(f"WORK EXPERIENCE:\n{exp_text[:800]}")
    
    # Add skills
    if resume_sections['skills']:
        skills_text = '\n'.join(resume_sections['skills'][:8])
        optimized_resume.append(f"SKILLS:\n{skills_text[:400]}")
    
    # Add education
    if resume_sections['education']:
        edu_text = '\n'.join(resume_sections['education'][:5])
        optimized_resume.append(f"EDUCATION:\n{edu_text[:200]}")
    
    optimized_resume_text = '\n\n'.join(optimized_resume)
    
    # Ensure we're within token limits
    resume_tokens = estimate_tokens(optimized_resume_text)
    if resume_tokens > max_resume_tokens:
        # Truncate if still too long
        chars_per_token = len(optimized_resume_text) / resume_tokens
        max_chars = int(max_resume_tokens * chars_per_token)
        optimized_resume_text = optimized_resume_text[:max_chars] + "... [truncated]"
    
    # Optimize job description
    job_lines = job_description.split('\n')
    important_lines = []
    
    for line in job_lines:
        line_lower = line.lower()
        # Prioritize lines with key information
        if any(keyword in line_lower for keyword in [
            'require', 'must', 'essential', 'experience', 'skill', 
            'qualification', 'bachelor', 'master', 'year', 'certification'
        ]):
            important_lines.append(line)
        elif line.strip() and len(important_lines) < 20:
            important_lines.append(line)
    
    optimized_job = '\n'.join(important_lines)
    
    # Ensure job description is within limits
    job_tokens = estimate_tokens(optimized_job)
    if job_tokens > max_job_tokens:
        chars_per_token = len(optimized_job) / job_tokens
        max_chars = int(max_job_tokens * chars_per_token)
        optimized_job = optimized_job[:max_chars] + "... [truncated]"
    
    return optimized_resume_text, optimized_job

def create_consistency_hash(resume_text, job_description, prompt_type):
    """Create a hash for identical inputs to ensure consistent outputs"""
    content = f"{resume_text[:1000]}{job_description[:1000]}{prompt_type}"
    return hashlib.md5(content.encode()).hexdigest()

def get_available_model(force_primary=True):
    """Get the first available model from the fallback chain"""
    if force_primary:
        # Force use of primary model without testing
        return MODEL_FALLBACK_CHAIN[0]  # gemini-2.5-flash
    
    for model in MODEL_FALLBACK_CHAIN:
        try:
            st.info(f"🔍 Testing model: {model}")
            
            test_model = genai.GenerativeModel(
                model,
                safety_settings={
                    genai.types.HarmCategory.HARM_CATEGORY_HARASSMENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
                    genai.types.HarmCategory.HARM_CATEGORY_HATE_SPEECH: genai.types.HarmBlockThreshold.BLOCK_NONE,
                    genai.types.HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: genai.types.HarmBlockThreshold.BLOCK_NONE,
                    genai.types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
                }
            )
            
            # Test with a simple prompt
            test_response = test_model.generate_content(
                "Say 'OK'",
                generation_config=genai.types.GenerationConfig(
                    temperature=0.1,
                    max_output_tokens=10
                )
            )
            
            if test_response.text:
                st.success(f"✅ {model} - Test successful!")
                return model
            else:
                st.warning(f"⚠️ {model} - No response text")
                
        except Exception as e:
            st.error(f"❌ {model} - Failed: {str(e)}")
            continue
    
    raise Exception("No available Gemini models found")

@rate_limit(min_interval=2)
def get_consistent_gemini_response(model_id, prompt, pdf_content, input_text, consistency_hash):
    """Enhanced response generation with robust error handling"""
    try:
        model = genai.GenerativeModel(
            model_id,
            safety_settings={
                genai.types.HarmCategory.HARM_CATEGORY_HARASSMENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
                genai.types.HarmCategory.HARM_CATEGORY_HATE_SPEECH: genai.types.HarmBlockThreshold.BLOCK_NONE,
                genai.types.HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: genai.types.HarmBlockThreshold.BLOCK_NONE,
                genai.types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: genai.types.HarmBlockThreshold.BLOCK_NONE,
            }
        )
        
        # Add consistency instruction to prompt
        enhanced_prompt = f"""
{prompt}

**CONSISTENCY PROTOCOL ACTIVE:**
Session ID: {consistency_hash}
Evaluation Date: {datetime.now().strftime('%Y-%m-%d')}
Apply identical methodology and scoring for consistent results.
Use deterministic analysis patterns and standardized language.

**RESUME CONTENT:**
{pdf_content[:3000]}

**JOB DESCRIPTION:**
{input_text[:2000]}
"""
        
        response = model.generate_content(
            enhanced_prompt,
            generation_config=genai.types.GenerationConfig(**GENERATION_CONFIG)
        )
        
        # Enhanced error checking
        if hasattr(response, 'candidates') and response.candidates:
            candidate = response.candidates[0]
            
            # Check finish reason
            if hasattr(candidate, 'finish_reason'):
                if candidate.finish_reason == 1:  # STOP - Normal completion
                    return response.text if hasattr(response, 'text') and response.text else "Analysis completed but no content returned."
                elif candidate.finish_reason == 2:  # MAX_TOKENS
                    return "⚠️ Analysis truncated due to length. Please try with a shorter resume or job description."
                elif candidate.finish_reason == 3:  # SAFETY
                    return "⚠️ Content filtered for safety. Please review your input for any potentially problematic content."
                elif candidate.finish_reason == 4:  # RECITATION
                    return "⚠️ Content blocked due to recitation concerns. Please try rephrasing your input."
                else:
                    return f"⚠️ Generation stopped with reason: {candidate.finish_reason}"
            
            # Try to get text anyway
            try:
                return response.text if response.text else "No analysis content generated."
            except:
                return "Analysis completed but content could not be retrieved."
        
        return "No response candidates generated. Please try again."
        
    except Exception as e:
        st.error(f"⚠️ Analysis Error: {str(e)}")
        
        # Fallback with simpler model configuration
        try:
            simple_model = genai.GenerativeModel("gemini-pro")
            simple_prompt = f"Analyze this resume against the job description:\n\nResume: {pdf_content[:1000]}\n\nJob: {input_text[:1000]}"
            
            fallback_response = simple_model.generate_content(simple_prompt)
            return f"⚠️ Using fallback analysis:\n\n{fallback_response.text}"
        except:
            return "Unable to complete analysis. Please check your API key, reduce content length, and try again."

def clean_extracted_text(text):
    """Clean and format extracted text"""
    # Remove excessive whitespace
    text = re.sub(r'\n\s*\n\s*\n', '\n\n', text)
    text = re.sub(r'[ \t]+', ' ', text)
    
    # Fix common extraction issues
    text = re.sub(r'([a-z])([A-Z])', r'\1 \2', text)  # Add space before capitals
    text = re.sub(r'(\w)([•·▪▫])', r'\1 \2', text)  # Space before bullets
    text = re.sub(r'([•·▪▫])(\w)', r'\1 \2', text)  # Space after bullets
    
    # Remove page markers
    text = re.sub(r'--- Page \d+ ---', '', text)
    
    # Normalize line endings
    text = text.replace('\r\n', '\n').replace('\r', '\n')
    
    # Remove empty lines at start and end
    text = text.strip()
    
    return text

def enhanced_pdf_processing(pdf_docs):
    """Enhanced PDF processing with better text extraction and formatting"""
    text = ""
    
    for doc in pdf_docs:
        try:
            if doc.name.endswith(".pdf"):
                # Save uploaded file temporarily
                with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_file:
                    tmp_file.write(doc.getvalue())
                    tmp_path = tmp_file.name
                
                try:
                    # Try multiple extraction methods
                    pdf_reader = PdfReader(tmp_path)
                    extracted_text = ""
                    
                    for page_num, page in enumerate(pdf_reader.pages):
                        page_text = page.extract_text()
                        
                        # Clean up common PDF extraction issues
                        page_text = re.sub(r'\s+', ' ', page_text)  # Normalize whitespace
                        page_text = re.sub(r'([a-z])([A-Z])', r'\1 \2', page_text)  # Add spaces between words
                        
                        extracted_text += f"\n--- Page {page_num + 1} ---\n{page_text}\n"
                    
                    # If extraction is poor, try alternative method
                    if len(extracted_text.strip()) < 100 and HAS_PDFPLUMBER:
                        try:
                            with pdfplumber.open(tmp_path) as pdf:
                                for page in pdf.pages:
                                    page_text = page.extract_text()
                                    if page_text:
                                        extracted_text += page_text + "\n"
                        except Exception:
                            pass
                    
                    text += extracted_text
                    
                finally:
                    # Clean up temporary file
                    os.unlink(tmp_path)
                    
            elif doc.name.endswith(".docx") and HAS_DOCX:
                try:
                    # Save uploaded file temporarily
                    with tempfile.NamedTemporaryFile(delete=False, suffix=".docx") as tmp_file:
                        tmp_file.write(doc.getvalue())
                        tmp_path = tmp_file.name
                    
                    try:
                        doc_reader = docx.Document(tmp_path)
                        
                        # Extract paragraphs
                        for para in doc_reader.paragraphs:
                            if para.text.strip():
                                text += para.text + "\n"
                        
                        # Extract tables
                        for table in doc_reader.tables:
                            for row in table.rows:
                                row_text = " | ".join([cell.text.strip() for cell in row.cells])
                                if row_text.strip():
                                    text += row_text + "\n"
                    
                    finally:
                        os.unlink(tmp_path)
                        
                except Exception as e:
                    st.error(f"📋 Error processing DOCX {doc.name}: {str(e)}")
                    
        except Exception as e:
            st.error(f"📄 Error processing {doc.name}: {str(e)}")
            continue
    
    # Clean and format the extracted text
    text = clean_extracted_text(text)
    return text

def validate_resume_content(text):
    """Validate that the extracted text looks like a resume"""
    text_lower = text.lower()
    
    # Check for common resume indicators
    resume_indicators = [
        'experience', 'education', 'skills', 'work', 'employment',
        'university', 'college', 'degree', 'certification', 'project',
        'email', 'phone', 'address', 'linkedin'
    ]
    
    found_indicators = sum(1 for indicator in resume_indicators if indicator in text_lower)
    
    if found_indicators < 3:
        st.warning("⚠️ The uploaded file may not be a resume. Please verify the content.")
        return False
    
    if len(text.strip()) < 200:
        st.warning("⚠️ The extracted text seems too short. Please check your file.")
        return False
    
    return True

def validate_configuration():
    """Validate system configuration"""
    issues = []
    
    # Check API key
    if not os.getenv("GOOGLE_API_KEY") and not st.session_state.get("api_key"):
        issues.append("❌ Google API Key not configured")
    
    # Check optional packages
    if not HAS_DOCX:
        issues.append("⚠️ Optional: Install python-docx for better DOCX support (pip install python-docx)")
    
    if not HAS_PDFPLUMBER:
        issues.append("⚠️ Optional: Install pdfplumber for better PDF extraction (pip install pdfplumber)")
    
    if not HAS_TIKTOKEN:
        issues.append("⚠️ Optional: Install tiktoken for better token estimation (pip install tiktoken)")
    
    return issues

@st.cache_data
def perform_enhanced_analysis(resume_text, job_description, analysis_type, custom_query=None, force_primary=True):
    """Main analysis function with all improvements"""
    
    # Initialize cache
    init_cache()
    
    # Optimize content length
    optimized_resume, optimized_job = optimize_content_length(resume_text, job_description)
    
    # Create consistency hash
    consistency_hash = create_consistency_hash(optimized_resume, optimized_job, analysis_type)
    
    # Try to get from cache first
    model_id = get_available_model(force_primary=force_primary)
    cached_response = get_cached_response(consistency_hash, model_id)
    
    if cached_response:
        st.success("⚡ Retrieved from cache for consistency")
        return cached_response, consistency_hash
    
    # Select prompt
    prompt_map = {
        "evaluate": "evaluate_resume",
        "improve": "improve_skills", 
        "keywords": "missing_keywords",
        "match": "percentage_match",
        "executive": "executive_assessment",
        "transition": "career_transition",
        "custom": "answer_query"
    }
    
    base_prompt = SPECIALIZED_PROMPTS[prompt_map.get(analysis_type, "evaluate_resume")]
    
    if analysis_type == "custom" and custom_query:
        base_prompt = f"{base_prompt}\n\nSPECIFIC QUERY: {custom_query}"
    
    # Generate response
    response = get_consistent_gemini_response(
        model_id, base_prompt, optimized_resume, optimized_job, consistency_hash
    )
    
    # Cache the response
    if response and not response.startswith("⚠️"):
        cache_response(consistency_hash, response, model_id)
    
    return response, consistency_hash

# Streamlit App Configuration
st.set_page_config(
    page_title="Smart ATS System", 
    page_icon="🚀",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Custom CSS for professional UI
st.markdown("""
<style>
    .main-header {
        background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
        padding: 2rem;
        border-radius: 15px;
        text-align: center;
        color: white;
        margin-bottom: 2rem;
        box-shadow: 0 10px 30px rgba(0,0,0,0.1);
    }
    .feature-card {
        background: linear-gradient(135deg, #f8f9fa 0%, #e9ecef 100%);
        padding: 1.5rem;
        border-radius: 12px;
        border-left: 5px solid #667eea;
        margin: 1rem 0;
        box-shadow: 0 5px 15px rgba(0,0,0,0.08);
    }
    .metric-container {
        background: linear-gradient(135deg, #28a745 0%, #20c997 100%);
        color: white;
        padding: 1.5rem;
        border-radius: 12px;
        text-align: center;
        margin: 1rem 0;
    }
    .consistency-badge {
        background: #ffc107;
        color: #212529;
        padding: 0.5rem 1rem;
        border-radius: 20px;
        font-weight: bold;
        display: inline-block;
        margin: 0.5rem 0;
    }
    .stButton > button {
        height: 3rem;
        font-weight: 600;
    }
</style>
""", unsafe_allow_html=True)

# Main Header
st.markdown("""
<div class="main-header">
    <h1>🚀 Smart ATS System</h1>
    <p>AI-Powered Resume Analysis with Guaranteed Consistency | Global Multi-Domain Support</p>
    <div class="consistency-badge">✅ Consistent Results Guaranteed</div>
</div>
""", unsafe_allow_html=True)



# Sidebar Configuration
with st.sidebar:
    st.markdown("### 🔑 Configuration")
    st.markdown("[Get your Google API Key](https://aistudio.google.com/app/apikey)")
    
    api_key = st.text_input("🔐 Google API Key", type="password", help="Your Gemini API key for AI analysis")
    st.session_state["api_key"] = api_key
    
    # Model selection option
    force_primary = st.checkbox("🎯 Force Primary Model", value=True, help="Use gemini-2.5-flash directly without testing")
    st.session_state["force_primary"] = force_primary
    
    if api_key:
        try:
            genai.configure(api_key=api_key)
            model_id = get_available_model(force_primary=force_primary)
            st.success(f"✅ Connected to {model_id}")
        except Exception as e:
            st.error(f"❌ API Key Error: {str(e)}")
    
    st.markdown("### 📂 Document Upload")
    uploaded_files = st.file_uploader(
        "📄 Upload Resume (PDF/DOCX)", 
        type=["pdf", "docx"], 
        accept_multiple_files=True,
        help="Upload your resume in PDF or DOCX format"
    )
    
    if uploaded_files:
        st.success(f"✅ {len(uploaded_files)} file(s) uploaded successfully!")
        

# Main Interface
st.markdown("### 📝 Job Description Input")
input_text = st.text_area(
    "Paste the complete job description here:",
    height=150,
    placeholder="Enter the full job description including requirements, responsibilities, and qualifications...",
    help="The more detailed the job description, the more accurate the analysis"
)

# Analysis Buttons
st.markdown("### 🎯 Choose Analysis Type")
col1, col2, col3 = st.columns(3)

with col1:
    evaluate_btn = st.button("📊 Complete Evaluation", use_container_width=True)
    improve_btn = st.button("📈 Skill Enhancement", use_container_width=True)

with col2:
    keywords_btn = st.button("🔍 Keyword Analysis", use_container_width=True)
    match_btn = st.button("🎯 Match Percentage", use_container_width=True)

with col3:
    executive_btn = st.button("👔 Executive Assessment", use_container_width=True)
    transition_btn = st.button("🔄 Career Transition", use_container_width=True)

# Custom Query Section
with st.expander("💬 Ask Custom Question"):
    custom_query = st.text_input("Ask any specific question about the resume or career advice:")
    query_btn = st.button("🤔 Get Expert Answer")

# Analysis Logic
analysis_triggered = any([evaluate_btn, improve_btn, keywords_btn, match_btn, executive_btn, transition_btn, query_btn])

if analysis_triggered:
    # Validation
    errors = []
    if not api_key:
        errors.append("🔐 Please enter your Google API Key")
    if not uploaded_files:
        errors.append("📄 Please upload your resume")
    if not input_text:
        errors.append("📝 Please paste the job description")
    
    if errors:
        for error in errors:
            st.error(error)
    else:
        # Process analysis
        with st.spinner("🔄 Analyzing with advanced AI algorithms..."):
            pdf_content = enhanced_pdf_processing(uploaded_files)
            
            # Validate content
            if not validate_resume_content(pdf_content):
                st.warning("⚠️ Please verify that your uploaded file is a valid resume.")
            
            # Determine analysis type
            if evaluate_btn:
                analysis_type = "evaluate"
            elif improve_btn:
                analysis_type = "improve"
            elif keywords_btn:
                analysis_type = "keywords"
            elif match_btn:
                analysis_type = "match"
            elif executive_btn:
                analysis_type = "executive"
            elif transition_btn:
                analysis_type = "transition"
            elif query_btn:
                analysis_type = "custom"
            
            try:
                # Get force_primary setting
                force_primary_setting = st.session_state.get("force_primary", True)
                
                # Perform analysis
                response, consistency_hash = perform_enhanced_analysis(
                    pdf_content, input_text, analysis_type, custom_query, force_primary_setting
                )
                
                # Display results
                st.markdown("## 📋 Analysis Results")
                
                # Show metadata
                col1, col2 = st.columns(2)
                with col1:
                    st.markdown(f"**Consistency ID:** `{consistency_hash[:8]}`")
                with col2:
                    st.markdown(f"**Analysis Type:** {analysis_type.title()}")
                
                st.markdown("---")
                st.markdown(response)
                
            
                
                # Show content optimization info
                if st.checkbox("🔍 Show Content Optimization Details"):
                    optimized_resume, optimized_job = optimize_content_length(pdf_content, input_text)
                    
                    col1, col2 = st.columns(2)
                    with col1:
                        st.metric("Resume Tokens", estimate_tokens(optimized_resume))
                        st.metric("Original Resume Length", len(pdf_content))
                    with col2:
                        st.metric("Job Description Tokens", estimate_tokens(optimized_job))
                        st.metric("Original Job Length", len(input_text))
                
            except Exception as e:
                st.error(f"Analysis failed: {str(e)}")
                st.info("Please try again with a shorter document or check your API key.")

# Footer
st.markdown("---")
st.markdown("""
<div style="text-align: center; color: #666;">
    <p>🚀 Smart ATS System  | Powered by Advanced AI | Consistent • Reliable • Universal</p>
    <p>Built with cutting-edge strategies for maximum ATS compatibility and career success</p>
</div>
""", unsafe_allow_html=True)

# Initialize cache on startup
if __name__ == "__main__":
    init_cache()