🧠 JobSense - Job Role Generation & Salary Prediction using Transformer

JobSense is an AI-powered job analysis tool that uses a GPT-style Transformer model to generate new job roles and predict corresponding salary ranges. The model is trained on real-world job data and aims to predict trends in employment, useful for career analysts, job seekers, and HR departments.

---

📌 Features

• 🔮 Generate realistic and unique job roles
• 💸 Predict average salary for generated job roles
• ⚙️ Custom Transformer model with PyTorch
• 📊 Dataset pre-processing for job roles and salary ranges
• 📈 Epoch-wise training loss tracking

---

📁 Project Structure

JobSense/
│
├── JobSense.ipynb           # Main notebook (model training, generation)
├── requirements.txt         # List of dependencies (auto-created below)
├── best_model.pth           # Saved transformer model weights (not included)
├── README.md                # You're reading this!

📂 Dataset

This project uses the Job Description Dataset by Ravindra Singh Rana, hosted on Kaggle.

🧾 Dataset Overview

• 📄 Filename: Job_Description_Dataset.csv
• 🧠 Features:
  • Job Title
  • Job Description
  • Key Skills
  • Role Category
  • Functional Area
  • Industry
  • Role
  • Salary
• 🔍 Used columns in this project:
  • Role (used to generate new job titles)
  • Salary (used to predict average salary)

📌 Example Entry

Role	Salary
Software Development Manager	₹15,00,000
Data Scientist	₹12,00,000
AI Research Engineer	₹20,00,000

---

🔍 Code Walkthrough

The notebook performs the following steps:

🔹 Data Loading & Preprocessing

• Loads a dataset with job Role and Salary Range.
• Cleans salary range values and converts them to a numerical average.
• Builds a character-level vocabulary for tokenization.

🔹 Model Definition

• Defines a custom Transformer using PyTorch’s nn.Transformer.
• Uses embedding layers, positional encoding, and a generator head.
• Creates two heads: one for job role generation, and one for salary prediction.

🔹 Training Loop

• Uses cross-entropy loss for the character prediction task.
• Trains the model using the Adam optimizer.
• Tracks loss and training time per epoch.
• Trains over 100 epochs for optimal performance.

🔹 Generation

• Performs autoregressive sampling from the model to generate new job titles.
• Predicts the average salary for the generated role using a separate regression head.

🔹 Model Saving

• Saves the trained model in best_model.pth (ensure you save this manually — not included in the repo).

---

⚙️ Installation

✅ Python ≥ 3.8 is recommended

📦 1. Clone the repository

git clone https://github.com/yourusername/JobSense.git
cd JobSense

📦 2. Install dependencies

pip install -r requirements.txt

If requirements.txt is not present, install manually:

pip install torch numpy pandas matplotlib scikit-learn

📌 Dependencies Below is the content of requirements.txt:

torch
numpy
pandas
matplotlib
scikit-learn

💡 Future Work

✅ Integrate job description generation using NLP

✅ Upgrade tokenizer to word-level or subword-level for higher accuracy

✅ Deploy as a Flask or Streamlit web app

✅ Integrate with real-time job APIs (e.g., LinkedIn, Naukri, etc.)