# ============================================================
# app.py — TalentMatch AI on Hugging Face Spaces
# This file runs the web interface using Gradio.
# Hugging Face automatically runs this file when your Space loads.
# ============================================================

import gradio as gr
import pandas as pd
from preprocess import full_preprocess
from skill_extractor import extract_skills, skill_match_percentage
from matcher import rank_candidates

# ── Load and preprocess data when the app FIRST starts ──────
# This runs ONCE when someone opens the Space — then it's ready.
print("Loading dataset...")
df = pd.read_csv("data/ats_resume_dataset_elite_v3.csv")

print("Preprocessing all resumes (this takes ~30 seconds)...")
df['cleaned_resume'] = df['resume_text'].apply(full_preprocess)

print("TalentMatch AI is ready!")


# ── The main function that Gradio calls ──────────────────────
def screen_candidates(job_description, num_results):
    """
    Takes a job description and returns the top matching candidates.
    This function is called every time someone clicks 'Submit'.
    """
    # Handle empty input
    if not job_description.strip():
        return pd.DataFrame({"Message": ["Please paste a job description in the box above."]})

    # Extract skills from the job description
    job_skills = extract_skills(job_description)

    # Rank all candidates using TF-IDF + Cosine Similarity
    top_candidates = rank_candidates(df, job_description, top_n=int(num_results))

    # Build a clean results table
    results = []
    for rank, (_, row) in enumerate(top_candidates.iterrows(), start=1):
        resume_skills = extract_skills(str(row['resume_text']))
        match_pct = skill_match_percentage(resume_skills, job_skills)

        results.append({
            'Rank'          : rank,
            'Resume ID'     : row['resume_id'],
            'Job Role'      : row['job_role'],
            'Experience'    : f"{row['experience_years']} yrs",
            'Education'     : row['education_level'],
            'ATS Score'     : round(row['computed_similarity'], 3),
            'Skill Match'   : f"{int(match_pct * 100)}%",
            'Shortlisted'   : 'YES ✅' if row['shortlisted'] == 1 else 'NO',
        })

    return pd.DataFrame(results)


# ── Build the Gradio web interface ───────────────────────────
demo = gr.Interface(
    fn=screen_candidates,

    inputs=[
        gr.Textbox(
            label="Paste Job Description Here",
            placeholder=(
                "Example:\n"
                "We are looking for a Data Scientist.\n"
                "Required skills: Python, Machine Learning, SQL, pandas, scikit-learn.\n"
                "Minimum 2 years experience required."
            ),
            lines=7
        ),
        gr.Slider(
            minimum=5,
            maximum=20,
            value=10,
            step=1,
            label="Number of Top Candidates to Show"
        ),
    ],

    outputs=gr.Dataframe(
        label="Top Matching Candidates — Ranked by AI",
        wrap=True
    ),

    title="TalentMatch AI — ATS Resume Screening System",
    description=(
        "Built using NLP, TF-IDF Vectorization, and Cosine Similarity. "
        "Paste any job description → AI ranks 6,000 resumes in seconds. "
        "Built by an Undergraduate CS Student."
    ),

    examples=[
        [
            "We need a Data Scientist with Python, Machine Learning, TensorFlow, SQL, and pandas. "
            "2+ years experience required. Knowledge of NLP and deep learning is a plus.",
            10
        ],
        [
            "Looking for a Frontend Developer skilled in React, JavaScript, HTML, CSS, and Node.js. "
            "Must know REST APIs and Git. Fresh graduates welcome.",
            8
        ],
        [
            "Hiring ML Engineer with PyTorch, scikit-learn, Docker, AWS, and Python. "
            "Must have experience with model deployment and feature engineering.",
            10
        ],
    ],

    theme=gr.themes.Soft(),

)

# ── Launch the app ────────────────────────────────────────────
# On Hugging Face, demo.launch() with no arguments is correct.
# Locally, use demo.launch(share=True) to get a public link.
demo.launch()