app.py

import os
import re
from datetime import datetime
import PyPDF2
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoModelForSeq2SeqLM
from sentence_transformers import SentenceTransformer, util
from groq import Groq
import gradio as gr
from docxtpl import DocxTemplate

# Set your API key for Groq
os.environ["GROQ_API_KEY"] = "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"
client = Groq(api_key=os.environ.get("GROQ_API_KEY"))

# --- PDF/Text Extraction Functions --- #
def extract_text_from_file(file_path):
    """Extracts text from PDF or TXT files based on file extension."""
    if file_path.endswith('.pdf'):
        return extract_text_from_pdf(file_path)
    elif file_path.endswith('.txt'):
        return extract_text_from_txt(file_path)
    else:
        raise ValueError("Unsupported file type. Only PDF and TXT files are accepted.")

def extract_text_from_pdf(pdf_file_path):
    """Extracts text from a PDF file."""
    with open(pdf_file_path, 'rb') as pdf_file:
        pdf_reader = PyPDF2.PdfReader(pdf_file)
        text = ''.join(page.extract_text() for page in pdf_reader.pages if page.extract_text())
    return text

def extract_text_from_txt(txt_file_path):
    """Extracts text from a .txt file."""
    with open(txt_file_path, 'r', encoding='utf-8') as txt_file:
        return txt_file.read()

# --- Skill Extraction with Llama Model --- #
def extract_skills_llama(text):
    """Extracts skills from the text using the Llama model via Groq API."""
    try:
        response = client.chat.completions.create(
            messages=[{"role": "user", "content": f"Extract skills from the following text: {text}"}],
            model="llama3-70b-8192",
        )
        skills = response.choices[0].message.content.split(', ')  # Expecting a comma-separated list
        return skills
    except Exception as e:
        raise RuntimeError(f"Error during skill extraction: {e}")

# --- Job Description Processing Function --- #
def process_job_description(text):
    """Extracts skills or relevant keywords from the job description."""
    return extract_skills_llama(text)

# --- Qualification and Experience Extraction --- #
def extract_qualifications(text):
    """Extracts qualifications from text (e.g., degrees, certifications)."""
    qualifications = re.findall(r'(bachelor|master|phd|certified|degree)', text, re.IGNORECASE)
    return qualifications if qualifications else ['No specific qualifications found']

def extract_experience(text):
    """Extracts years of experience from the text."""
    experience_years = re.findall(r'(\d+)\s*(years|year) of experience', text, re.IGNORECASE)
    job_titles = re.findall(r'\b(software engineer|developer|manager|analyst)\b', text, re.IGNORECASE)
    experience_years = [int(year[0]) for year in experience_years]
    return experience_years, job_titles

# --- Summarization Function --- #
def summarize_experience(experience_text):
    """Summarizes the experience text using a pre-trained model."""
    model_name = "facebook/bart-large-cnn"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

    inputs = tokenizer(experience_text, return_tensors="pt", max_length=1024, truncation=True)
    summary_ids = model.generate(inputs['input_ids'], max_length=150, min_length=30, length_penalty=2.0, num_beams=4, early_stopping=True)
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary

# --- Semantic Similarity Calculation --- #
def calculate_semantic_similarity(text1, text2):
    """Calculates semantic similarity using a sentence transformer model and returns the score as a percentage."""
    model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
    embeddings1 = model.encode(text1, convert_to_tensor=True)
    embeddings2 = model.encode(text2, convert_to_tensor=True)
    similarity_score = util.pytorch_cos_sim(embeddings1, embeddings2).item()
    
    # Convert similarity score to percentage
    similarity_percentage = similarity_score * 100
    return similarity_percentage

# --- Thresholds --- #
def categorize_similarity(score):
    """Categorizes the similarity score into thresholds for better insights."""
    if score >= 80:
        return "High Match"
    elif score >= 50:
        return "Moderate Match"
    else:
        return "Low Match"

# --- Communication Generation with Enhanced Response --- #
def communication_generator(resume_skills, job_description_skills, skills_similarity, qualifications_similarity, experience_similarity, candidate_experience):
    """Generates a detailed communication response based on similarity scores and additional criteria."""
    # Assess candidate fit based on similarity scores
    fit_status = "strong fit" if skills_similarity >= 80 and qualifications_similarity >= 80 and experience_similarity >= 80 else \
                 "moderate fit" if skills_similarity >= 50 else "weak fit"
    
    # Build a message that includes a recommendation based on various assessments
    if fit_status == "strong fit":
        recommendation = "We recommend moving forward with this candidate, as they demonstrate a high level of alignment with the role requirements."
    elif fit_status == "moderate fit":
        recommendation = "This candidate shows potential; however, further assessment or interviews are recommended to clarify their fit for the role."
    else:
        recommendation = "We advise against moving forward with this candidate, as they do not meet the key technical requirements for the position."

    message = (
        f"After a detailed analysis of the candidate's resume, we found the following insights:\n\n"
        f"- **Skills Match**: {skills_similarity:.2f}% (based on required technologies: {', '.join(job_description_skills)})\n"
        f"- **Experience Match**: {experience_similarity:.2f}% (relevant experience: {candidate_experience} years)\n"
        f"- **Qualifications Match**: {qualifications_similarity:.2f}%\n\n"
        f"The overall assessment indicates that the candidate is a {fit_status} for the role. "
        f"Their skills in {', '.join(resume_skills)} align with the job's requirements of {', '.join(job_description_skills)}. "
        f"Based on their experience in web application development, particularly with technologies like {', '.join(resume_skills)}, they could contribute effectively to our team.\n\n"
        f"**Recommendation**: {recommendation}\n"
    )

    return message

# --- Sentiment Analysis --- #
def sentiment_analysis(text):
    """Analyzes the sentiment of the text."""
    model_name = "mrm8488/distiluse-base-multilingual-cased-v2-finetuned-stsb_multi_mt-es"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)

    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
    with torch.no_grad():
        outputs = model(**inputs)
        predicted_sentiment = torch.argmax(outputs.logits).item()
    return ["Negative", "Neutral", "Positive"][predicted_sentiment]

# --- Updated Resume Analysis Function --- #
def analyze_resume(resume_file, job_description_file):
    """Analyzes the resume and job description, returning similarity score, skills, qualifications, and experience matching."""
    # Extract resume and job description text
    try:
        resume_text = extract_text_from_file(resume_file.name)
        job_description_text = extract_text_from_file(job_description_file.name)
    except ValueError as ve:
        return str(ve)

    # Extract skills, qualifications, and experience
    resume_skills = extract_skills_llama(resume_text)
    job_description_skills = process_job_description(job_description_text)
    resume_qualifications = extract_qualifications(resume_text)
    job_description_qualifications = extract_qualifications(job_description_text)
    resume_experience, resume_job_titles = extract_experience(resume_text)
    job_description_experience, job_description_titles = extract_experience(job_description_text)

    # Summarize experiences
    resume_experience_summary = summarize_experience(resume_text)
    job_description_experience_summary = summarize_experience(job_description_text)

    # Calculate semantic similarity for different sections in percentages
    skills_similarity = calculate_semantic_similarity(' '.join(resume_skills), ' '.join(job_description_skills))
    qualifications_similarity = calculate_semantic_similarity(' '.join(resume_qualifications), ' '.join(job_description_qualifications))
    experience_similarity = calculate_semantic_similarity(' '.join([str(e) for e in resume_experience]), ' '.join([str(e) for e in job_description_experience]))

    # Assuming candidate experience is the total of years from the resume
    candidate_experience = sum(resume_experience)

    # Generate communication based on analysis
    response_message = communication_generator(
        resume_skills,
        job_description_skills,
        skills_similarity,
        qualifications_similarity,
        experience_similarity,
        candidate_experience
    )

    return {
        "Resume Experience Summary": resume_experience_summary,
        "Job Description Experience Summary": job_description_experience_summary,
        "Skills Similarity": skills_similarity,
        "Qualifications Similarity": qualifications_similarity,
        "Experience Similarity": experience_similarity,
        "Candidate Experience": candidate_experience,
        "Response Message": response_message,
    }

# --- Gradio Interface --- #
with gr.Blocks() as demo:
    gr.Markdown("## Resume and Job Description Analyzer")
    with gr.Row():
        resume_file_input = gr.File(label="Upload Resume (PDF/TXT)", type="file")
        job_description_file_input = gr.File(label="Upload Job Description (PDF/TXT)", type="file")
    analyze_button = gr.Button("Analyze")
    results_output = gr.Output()

    analyze_button.click(analyze_resume, inputs=[resume_file_input, job_description_file_input], outputs=results_output)

# Launch Gradio Interface
demo.launch()