TalentMiner

TalentMiner is an AI-powered resume screening platform that matches a candidate resume against a job description and returns:

• Match score (0-100)
• Match status (Good Match, Average Match, Poor Match)
• Predicted resume category
• Matched and missing skills
• Actionable improvement suggestions

It includes a Python Flask backend, a React + TypeScript frontend, and a machine learning pipeline for resume category classification.

Features

• Resume and job description analysis via API (/api/analysis)
• Resume upload flow with text extraction support in frontend
• ATS-style skill gap analysis (matched vs missing keywords)
• Weighted scoring from semantic similarity and skill-priority matching
• Resume category prediction using trained ML artifacts
• Optional semantic engine using SentenceTransformers (all-MiniLM-L6-v2)
• Auth flow: signup, login, logout, profile update, password change

Tech Stack

• Frontend: React, TypeScript, Vite, Tailwind CSS, shadcn/ui, React Router
• Backend: Flask, Flask-CORS, scikit-learn, joblib, Werkzeug auth
• Optional NLP/ATS extras: pypdf, sentence-transformers
• Model training: pandas, numpy, scikit-learn, nltk

Project Structure

TalentMiner/
  backend/                  # Flask API and auth/session logic
  dataset/                  # Resume.csv dataset
  frontend/
    hassan-code-canvas-main/  # React + Vite client app
  model/                    # Training scripts, inference check, model artifacts
  notebook/                 # Phase notebooks and reports
  PROJECT_REPORT.md         # Detailed project report

How Scoring Works

Final score combines:

• Semantic similarity between resume and JD
• Priority-weighted skill coverage from JD keyword ranking

Formula used in backend:

final_score = 0.55 * semantic_score + 0.45 * skill_score
score = clamp(final_score, 0, 1) * 100

Where skill priority is inferred from JD wording (for example: required, mandatory, preferred, bonus).

Model Details

Current model metadata (model/metadata.json):

• Best model: linear_svc
• Dataset rows used: 2484
• Classes: 24
• Vectorizer: TF-IDF (1,2)-grams, max_features=5000
• Dimensionality reduction: TruncatedSVD (100 components)
• Best macro F1: ~0.601

Artifacts expected in model/

• best_model.pkl
• tfidf_vectorizer.pkl
• label_encoder.pkl
• svd.pkl
• metadata.json

Setup and Run

1) Backend (Flask)

From project root:

python -m venv .venv
.venv\Scripts\activate
pip install -r backend/requirements.txt

Optional advanced ATS features:

pip install -r backend/requirements-ats-advanced.txt

Run backend:

python backend/app.py

Backend runs on:

• http://127.0.0.1:5000

Health check:

• GET http://127.0.0.1:5000/api/health

2) Frontend (React + Vite)

cd frontend/hassan-code-canvas-main
npm install
npm run dev

Frontend default dev URL:

• http://localhost:5173

API base URL is read from VITE_API_BASE_URL and defaults to:

• http://127.0.0.1:5000

You can create a .env in frontend/hassan-code-canvas-main:

VITE_API_BASE_URL=http://127.0.0.1:5000

API Endpoints

Health and analysis

• GET /api/health
• POST /api/analysis
  • Body: { "resumeText": string, "jobDescription": string }
• POST /api/analysis-upload
  • Multipart with resume file + job description

Auth and profile

• POST /api/auth/signup
• POST /api/auth/login
• POST /api/auth/logout
• GET /api/auth/me
• PUT /api/profile
• POST /api/change-password

Training and Inference

Install model dependencies:

pip install -r model/requirements.txt

Train and export artifacts:

python model/train_and_export.py

Run inference smoke check:

python model/inference_check.py

Optional custom text:

python model/inference_check.py --text "Experienced Python developer building REST APIs and ML pipelines"

Notes and Limitations

• Sessions/tokens are in-memory in backend and reset when server restarts.
• User persistence is JSON-file based (backend/users.json) and intended for demo/small-scale usage.
• BERT loading can increase startup time when enabled.
• For production, use a real database and a production WSGI server.

Roadmap Ideas

• JWT-based auth with refresh tokens and expiry
• Persistent database for users and analysis history
• Exportable PDF/CSV analysis reports
• Better confidence calibration and per-category explainability
• Automated regression and integration testing

References

• Full report: PROJECT_REPORT.md
• Model docs: model/README.md
• Phase notebooks: notebook/