import gradio as gr
import spacy
import pdfplumber
from docx import Document
from sentence_transformers import SentenceTransformer, util
import re
import plotly.graph_objects as go
import sys
import os
import torch

from transformers import pipeline

# Initialize Models once at startup
print("🚀 [1/4] Initializing Semantic Matcher (all-MiniLM-L6-v2)...")
try:
    model = SentenceTransformer('all-MiniLM-L6-v2', device='cpu')
except Exception as e:
    print(f"❌ Semantic Matcher failed: {e}")
    model = None

print("🚀 [2/4] Initializing Reasoning LLM (FLAN-T5-Base)...")
try:
    # Use CPU by default for stability on Windows unless explicitly requested
    llm_reasoner = pipeline("text2text-generation", model="google/flan-t5-base", device=-1)
except Exception as e:
    print(f"⚠️ Reasoning LLM failed: {e}")
    llm_reasoner = None

print("🚀 [3/4] Initializing Interview Coach (LaMini-GPT)...")
try:
    llm_coach = pipeline("text-generation", model="MBZUAI/LaMini-GPT-124M", device=-1)
except Exception as e:
    print(f"⚠️ Coach LLM failed: {e}")
    llm_coach = None

print("🚀 [4/4] Loading NLP Entities (spaCy)...")
try:
    nlp = spacy.load("en_core_web_sm")
except:
    print("📥 Downloading spaCy model...")
    os.system(f"{sys.executable} -m spacy download en_core_web_sm")
    try:
        nlp = spacy.load("en_core_web_sm")
    except:
        nlp = None
        print("❌ NLP Load failed completely.")

print("✨ Intelligence Engine Ready.")

TECH_SKILLS = [
    "python", "javascript", "react", "fastapi", "aws", "docker", "kubernetes", "sql", 
    "git", "ml", "nlp", "tensorflow", "pytorch", "java", "golang", "postgresql", 
    "mongodb", "redis", "devops", "rest api", "graphql", "scikit-learn", "pandas", 
    "numpy", "django", "flask", "typescript", "angular", "vue", "html", "css", 
    "node.js", "express", "azure", "gcp", "linux", "terraform", "ansible", "jenkins",
    "prompt engineering", "openai", "llm", "bert", "tableau", "powerbi", "excel",
    "c#", "c++", "rust", "r", "spark", "hadoop", "kafka", "mysql", "oracle", "snowflake"
]

ROADMAP_DB = {
    "python": "🐍 [Master Python](https://realpython.com/) - Focus on Backend Automation & Data Science.",
    "react": "⚛️ [React.dev](https://react.dev/) - Master Hooks & State Management.",
    "aws": "☁️ [AWS Builder](https://explore.skillbuilder.aws/) - Get Certified (Solutions Architect).",
    "docker": "🐳 [Docker Guide](https://docs.docker.com/) - Learn Container Architecture.",
    "kubernetes": "☸️ [K8s Certification](https://kubernetes.io/docs/tutorials/) - Master Orchestration.",
    "ml": "🤖 [ML Specialization](https://www.coursera.org/specializations/machine-learning-introduction) - Focus on Scikit-Learn.",
    "nlp": "✍️ [HF NLP Course](https://huggingface.co/learn/nlp-course/) - Master Transformers.",
    "sql": "💾 [SQL Practice](https://sqlzoo.net/) - Master Joins & Query Optmization.",
    "javascript": "📜 [JS.info](https://javascript.info/) - Master ES6+ Features.",
    "devops": "⚙️ [Roadmap.sh/devops](https://roadmap.sh/devops/) - Learn CI/CD & Infrastructure as Code."
}

def extract_text(file_obj):
    """Robust text extraction for PDF and DOCX."""
    if file_obj is None:
        return ""
    
    file_path = file_obj.name if hasattr(file_obj, 'name') else str(file_obj)
    
    try:
        if file_path.lower().endswith('.pdf'):
            with pdfplumber.open(file_path) as pdf:
                return "".join([page.extract_text() or "" for page in pdf.pages])
        elif file_path.lower().endswith('.docx'):
            doc = Document(file_path)
            return "\n".join([p.text for p in doc.paragraphs])
        else:
            # Try reading as plain text
            with open(file_path, 'r', encoding='utf-8') as f:
                return f.read()
    except Exception as e:
        print(f"Extraction error on {file_path}: {e}")
    return ""

def discover_skills(text):
    if not text: return set()
    found = set()
    text_l = text.lower()
    for skill in TECH_SKILLS:
        if re.search(r'\b' + re.escape(skill) + r'\b', text_l):
            found.add(skill)
    return found

def create_score_gauges(match_score, content_score, search_score, ats_score):
    def make_gauge(val, title, color):
        return go.Indicator(
            mode="gauge+number",
            value=val,
            title={'text': title, 'font': {'size': 14, 'color': "white"}},
            domain={'x': [0, 1], 'y': [0, 1]},
            gauge={
                'axis': {'range': [0, 100], 'tickwidth': 1, 'tickcolor': "white"},
                'bar': {'color': color},
                'bgcolor': "rgba(0,0,0,0)",
                'borderwidth': 2,
                'bordercolor': "gray",
                'steps': [
                    {'range': [0, 40], 'color': 'rgba(255, 0, 0, 0.1)'},
                    {'range': [40, 70], 'color': 'rgba(255, 255, 0, 0.1)'},
                    {'range': [70, 100], 'color': 'rgba(0, 255, 0, 0.1)'}
                ],
            }
        )

    fig = go.Figure()
    fig.add_trace(make_gauge(match_score, "Match Score", "#00dfd8"))
    
    fig.update_layout(
        paper_bgcolor='rgba(0,0,0,0)',
        plot_bgcolor='rgba(0,0,0,0)',
        font={'color': "white", 'family': "Arial"},
        height=300,
        margin=dict(l=40, r=40, t=80, b=40)
    )
    return fig

def create_radar_chart(skills, exp, edu, readiness, search):
    categories = ['Skills', 'Experience', 'Education', 'Readiness', 'Searchability']
    fig = go.Figure()

    fig.add_trace(go.Scatterpolar(
        r=[skills, exp, edu, readiness, search],
        theta=categories,
        fill='toself',
        name='Competency Profile',
        line_color='#00dfd8',
        fillcolor='rgba(0, 223, 216, 0.3)'
    ))

    fig.update_layout(
        polar=dict(
            radialaxis=dict(visible=True, range=[0, 100], color="white", gridcolor="gray"),
            angularaxis=dict(color="white", gridcolor="gray"),
            bgcolor='rgba(0,0,0,0)'
        ),
        showlegend=False,
        paper_bgcolor='rgba(0,0,0,0)',
        plot_bgcolor='rgba(0,0,0,0)',
        height=350,
        margin=dict(l=60, r=60, t=20, b=20)
    )
    return fig

def estimate_salary(score, skills):
    base = 75
    multiplier = 1 + (score / 100)
    skill_bonus = len(skills) * 2.5
    low = round(base * multiplier + skill_bonus)
    high = round(low * 1.45)
    return f"${low}k - ${high}k"

def main_process(resume_file, jd_text, progress=gr.Progress()):
    print("\n" + "="*40)
    print("🚀 INITIALIZING NEURAL INTELLIGENCE CYCLE")
    print("="*40)
    
    try:
        if not resume_file or not jd_text.strip():
            print("⚠️ Incomplete inputs detected.")
            return [
                "⚠️ Missing Inputs", "", None, None, 
                "Incomplete data sequence. Please upload a resume and paste the JD.", "N/A", [], gr.update(visible=False)
            ]

        progress(0.1, desc="⚡ Extracting Neural Content...")
        print("Stage 1: Text Extraction...")
        resume_text = extract_text(resume_file)
        if not resume_text.strip():
            print("❌ Extraction failure: text is empty.")
            return [
                "❌ Extract Error", "", None, None, 
                "Neural bypass failed. Ensure the resume is a readable PDF or DOCX (not just an image).", "N/A", [], gr.update(visible=False)
            ]

        progress(0.2, desc="🔍 Discovering Skill Entities...")
        print("Stage 2: Skill Discovery...")
        r_skills = discover_skills(resume_text)
        j_skills = discover_skills(jd_text)
        match_skills = sorted(list(r_skills.intersection(j_skills)))
        gap_skills = sorted(list(j_skills - r_skills))
        print(f"   - Identified {len(match_skills)} matches and {len(gap_skills)} gaps.")

        progress(0.4, desc="📐 Computing Semantic Distance...")
        print("Stage 3: Embedding Computation...")
        if not model:
            print("❌ Error: Semantic model not loaded.")
            return ["Model Load Error", "N/A", None, None, "The semantic matching engine failed to initialize. Try restarting.", "N/A", [], gr.update(visible=False)]
            
        emb1 = model.encode(resume_text, convert_to_tensor=True)
        emb2 = model.encode(jd_text, convert_to_tensor=True)
        score = round(util.pytorch_cos_sim(emb1, emb2).item() * 100, 1)
        print(f"   - Semantic Match Score: {score}%")

        progress(0.7, desc="🤖 Syncing Neural Consensus...")
        print("Stage 4: LLM Ensemble Reasoning...")
        ensemble_insight = []
        
        # Model 1: Reasoner (FLAN-T5)
        if llm_reasoner:
            try:
                print("   - Querying Reasoner (FLAN-T5)...")
                prompt_t5 = f"Analyze resume relevance. Score: {score}%. Gaps: {', '.join(gap_skills)}. Summarize fit."
                t5_out = llm_reasoner(prompt_t5, max_length=100)[0]['generated_text']
                ensemble_insight.append(f"**Reasoner**: {t5_out}")
            except Exception as e:
                print(f"   - T5 Inference Error: {e}")
        else:
            print("   - Reasoner LLM not available. Skipping.")
            ensemble_insight.append("**Reasoner**: Neural reasoning offline. Using heuristic fallback.")

        # Model 2: Coach (LaMini)
        if llm_coach:
            try:
                print("   - Querying Coach (LaMini)...")
                prompt_lamini = f"Career coach advice for {score}% match."
                lamini_out = llm_coach(prompt_lamini, max_length=100, truncation=True)[0]['generated_text']
                # Clean up if output contains prompt
                clean_lamini = lamini_out.split("match.")[-1].strip() if "match." in lamini_out else lamini_out
                ensemble_insight.append(f"**Coach**: {clean_lamini}")
            except Exception as e:
                print(f"   - LaMini Inference Error: {e}")
        else:
            print("   - Coach LLM not available. Skipping.")
            ensemble_insight.append("**Coach**: Coaching strategy offline.")

        ai_analysis = "\n\n".join(ensemble_insight) if ensemble_insight else f"Neural consensus reached alignment at {score}%."

        # Heuristic metrics
        print("Stage 5: Calculating Dashboard Metrics...")
        content_score = min(100, len(resume_text.split()) / 4) 
        search_score = min(100, len(r_skills) * 12)
        
        progress(0.9, desc="🎨 Rendering Compass Interface...")
        print("Stage 6: Finalizing UI Components...")
        gauge_plot = create_score_gauges(score, content_score, search_score, score*0.9)
        radar_plot = create_radar_chart(len(match_skills)*12, 80, 85, score, search_score)
        salary_range = estimate_salary(score, match_skills)
        
        present_str = ", ".join([s.upper() for s in match_skills]) if match_skills else "No direct matches."
        gap_str = ", ".join([s.upper() for s in gap_skills]) if gap_skills else "No critical gaps!"

        print(f"✅ CYCLE COMPLETE | Score: {score}%")
        # Return exactly what the UI elements expect in the correct order
        return [
            present_str, gap_str, gauge_plot, radar_plot, 
            ai_analysis, salary_range, gap_skills, gr.update(visible=True)
        ]
    except Exception as e:
        print(f"🔥 CRITICAL SYSTEM FAULT: {str(e)}")
        return [
            "⚠️ System Error", "Analysis Failed", None, None, 
            f"An unexpected error occurred: {str(e)}", "N/A", [], gr.update(visible=False)
        ]

def generate_roadmap(gap_skills):
    if not gap_skills:
        return """
        <div style='text-align: center; padding: 40px; background: rgba(0, 223, 216, 0.05); border-radius: 15px; border: 1px dashed #00dfd8;'>
            <h2 style='color: #00dfd8; margin: 0;'>🌟 Neural Consensus: 100% Ready</h2>
            <p style='color: #8b949e;'>Both Reasoner & Coach models agree: Your profile is industry-ready.</p>
        </div>
        """
    
    # Using LLMs to generate more specific roadmap tips if available
    llm_tips = ""
    if llm_coach:
        try:
            prompt = f"Provide a one-sentence learning goal for each skill: {', '.join(gap_skills)}."
            llm_tips = llm_coach(prompt, max_length=150)[0]['generated_text']
        except:
            llm_tips = "Focus on documentation and hands-on builds."

    cards_html = f"<div style='margin-bottom: 20px; color: #8b949e; font-size: 11px; font-style: italic;'>AI Insight: {llm_tips}</div>"
    cards_html += "<div style='display: grid; grid-template-columns: repeat(auto-fill, minmax(250px, 1fr)); gap: 20px;'>"
    for s in gap_skills:
        res = ROADMAP_DB.get(s.lower(), f"Master {s.upper()} via specialized projects.")
        cards_html += f"""
        <div style='background: rgba(255, 255, 255, 0.03); border: 1px solid rgba(0, 223, 216, 0.2); border-radius: 12px; padding: 15px;'>
            <div style='display: flex; justify-content: space-between;'>
                <span style='color: #00dfd8; font-size: 10px;'>NEURAL UPGRADE</span>
                <span style='color: #8b949e; font-size: 10px;'>LVL: EXPERT</span>
            </div>
            <h4 style='color: #fff; margin: 10px 0 5px 0;'>{s.upper()}</h4>
            <p style='color: #8b949e; font-size: 11px; margin-bottom: 12px;'>{res}</p>
        </div>
        """
    cards_html += "</div>"
    return f"<div>{cards_html}</div>"

def generate_interview_questions(gaps):
    print("🤖 Neural Ensemble Coaching active...")
    context = ", ".join(gaps[:3]) if gaps else "strategic leadership"
    
    # Model 1: Reasoner (FLAN-T5) generates formal questions
    q1 = "1. Walk us through a scenario where you bridged technical gaps."
    if llm_reasoner:
        try:
            prompt_t5 = f"List 2 technical questions for {context}."
            q1 = llm_reasoner(prompt_t5, max_length=100)[0]['generated_text']
        except: pass

    # Model 2: Coach (LaMini) generates behavioral tips
    tips = "Focus on your adaptability."
    if llm_coach:
        try:
            prompt_lamini = f"Give a career coaching tip for someone learning {context}."
            tips = llm_coach(prompt_lamini, max_length=100)[0]['generated_text']
        except: pass

    return f"### 🎤 Neural Ensemble Coaching\n**Technical Focus**: {q1}\n\n**Career Strategy**: {tips}\n\n*This insight was cross-verified by T5 and LaMini models.*"

# Premium CSS for Glassmorphism
STYLE = """
.gradio-container { background-color: #0b0e14 !important; color: white !important; font-family: 'Inter', sans-serif !important; }
.glass-panel { background: rgba(255, 255, 255, 0.03) !important; border: 1px solid rgba(255, 255, 255, 0.08) !important; border-radius: 12px !important; padding: 18px !important; }
.hub-tag { background: #1a1e26; border: 1px solid #30363d; border-radius: 6px; padding: 4px 10px; font-size: 11px; color: #8b949e; display: inline-block; margin-right: 5px; }
.active-tag { background: rgba(0, 223, 216, 0.1); border-color: #00dfd8; color: #00dfd8; }
"""

with gr.Blocks(theme=gr.themes.Soft(), css=STYLE, title="Career Compass AI") as demo:
    with gr.Row():
        with gr.Column(scale=3):
            gr.HTML("""
                <div style="padding: 10px;">
                    <h1 style="color: #00dfd8; margin: 0; font-size: 2.2em;">🧭 Career Compass</h1>
                    <p style="color: #8b949e; margin-top: 5px;">Unified Neural Ensemble Dashboard</p>
                </div>
            """)
        with gr.Column(scale=2):
            gr.HTML(f"""
                <div style="text-align: right; padding-top: 20px;">
                    <span class="hub-tag active-tag">Ensemble Mode: ACTIVE</span>
                    <span class="hub-tag">T5-Base</span>
                    <span class="hub-tag">LaMini-GPT</span>
                </div>
            """)

    with gr.Row():
        with gr.Column(scale=1, variant="panel"):
            gr.Markdown("### 📥 Neural Feed")
            resume_input = gr.File(label="Target Resume")
            jd_input = gr.Textbox(label="Requirement Set", lines=8, placeholder="Paste JD sequence...")
            run_btn = gr.Button("⚡ Run Intelligent Cycle", variant="primary")
        
        with gr.Column(scale=2):
            with gr.Row():
                with gr.Column():
                    gr.Markdown("### 📊 Alignment Core")
                    gauge_plot = gr.Plot(label="Match Quality")
                with gr.Column():
                    gr.Markdown("### 🕸️ Competency Radar")
                    radar_plot = gr.Plot(label="Skills Radar")
            
            with gr.Row():
                with gr.Column(elem_classes=["glass-panel"]):
                    gr.Markdown("### 💡 Ensemble Insight")
                    analysis_text = gr.Markdown("Waiting for neural consensus...")
                with gr.Column(elem_classes=["glass-panel"]):
                    gr.Markdown("### 💰 Market Value")
                    salary_display = gr.Textbox(label="Projected Range", interactive=False)

    with gr.Row():
        with gr.Column():
            match_display = gr.Textbox(label="Neural Matches", interactive=False)
        with gr.Column():
            gap_display = gr.Textbox(label="Identified Faults", interactive=False)

    with gr.Column(visible=False) as roadmap_container:
        gr.Markdown("---")
        with gr.Row(elem_classes=["glass-panel"]):
            with gr.Column(scale=2):
                gr.Markdown("### 🚀 Knowledge Upgrade Path")
                roadmap_btn = gr.Button("Initialize Pathfinder", variant="secondary")
                roadmap_output = gr.HTML()
            with gr.Column(scale=1):
                gr.Markdown("### 🎤 Interview Coaching")
                interview_prep = gr.Markdown("Pathfinder inactive.")

    gap_state = gr.State(value=[])

    run_btn.click(
        fn=main_process,
        inputs=[resume_input, jd_input],
        outputs=[
            match_display, gap_display, gauge_plot, radar_plot, 
            analysis_text, salary_display, gap_state, roadmap_container
        ]
    )

    roadmap_btn.click(
        fn=lambda gaps: [generate_roadmap(gaps), generate_interview_questions(gaps)],
        inputs=[gap_state],
        outputs=[roadmap_output, interview_prep]
    )

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