Resume Scorer

Job hunting is brutal. Resumes get lost, advice is vague. I built Resume Scorer an AI tool that analyzes resumes like a human, gives specific fixes, and uses transparent ML.

Comment

Inspiration

Job hunting is a nightmare. You tweak your resume endlessly, submit it, and... crickets. Why? Because most resumes get filtered out by automated systems (ATS) before a human even sees them. Generic advice like "add more keywords" or "improve formatting" isn’t helpful you need specific, actionable feedback tailored to your target role. I built Resume Scorer to solve this. No more guessing games. No vague tips. Just data-driven, role-specific recommendations to make your resume stand out whether you're applying for Software Engineering, Data Science, or Product Management.

What It Does

Resume Scorer is an AI-powered tool that:

  • Analyzes your resume like a hiring manager—extracting skills, experience, and structure.
  • Predicts the best roles for you based on your background (not just keywords).
  • Gives specific fixes—"Add AWS projects for DevOps roles" or "Quantify achievements in your experience section."
  • Compares against job descriptions to highlight skill gaps.

Example Output: "Your resume is 85% match for Data Scientist roles. Strengths: Python, SQL, ML. Weaknesses: No TensorFlow projects. Add 1-2 ML project links to boost your score to 92%."

How We Built It

  1. Data Collection & Preprocessing
    • Scraped 2,000+ real resumes (PDFs/DOCX) from public datasets and anonymized submissions.
    • Manually labeled them by role (Software Eng, Data Sci, etc.) and quality (Good/Bad).
    • Extracted structured data using PyPDF2, regex, and custom NLP pipelines (no off-the-shelf parsers).
  2. Machine Learning Models Role Prediction: Ensemble of Random Forest + XGBoost trained on:
    • Skill frequencies (e.g., Python: 12 mentions)
    • Experience metrics (e.g., 5 years in SWE)
    • Education level (e.g., Masters = +10% for Data Sci)
    • Resume Quality Scoring: Regression model (R² = 0.89) evaluating:
    • Structure (sections, readability)
    • Content (quantified achievements, keyword density)
    • ATS Compliance (font, headers, length)
  3. Job Matching Engine
    • Uses TF-IDF + Cosine Similarity to compare resumes against job descriptions.
    • Identifies missing skills (e.g., "Job requires Kubernetes; add it!").
  4. Web App (Flask + React)
    • Drag-and-drop resume upload → instant analysis.
    • Clean, interactive dashboard showing scores + fixes.

Challenges We Ran Into

  • PDF Parsing Hell – Resumes come in wild formats (tables, columns, weird fonts). Had to build custom text extraction with fallbacks.
  • Skill Taxonomy – Not all "Python" mentions are equal (e.g., "Used Python" vs. "Built Python API"). Created a weighted skill-scoring system.
  • Bias in Training Data – Most resumes were from tech. Mitigated by oversampling non-tech roles.

Accomplishments We’re Proud Of

  • 89% accuracy in role prediction (tested on 500 real resumes).
  • Real impact – Users reported 2x more interview callbacks after using our suggestions.
  • No LLMs – Everything is interpretable ML (no ChatGPT magic).

Built With

Share this project:

Updates