cmkarthi2004/personal-career-assistant

Personal Career Assistant

Overview

Personal Career Assistant is a small, custom conversational dataset created to fine-tune an instruction-following language model for answering questions about my education, technical skills, academic projects, freelance experience, client projects, and career interests.

The dataset is designed for a personal AI assistant prototype and contains question-and-answer examples in JSONL format.

Dataset Files

File	Description
career_dataset.jsonl	Training dataset containing conversational question-and-answer examples
personal_career_assistant_training.ipynb	Google Colab notebook containing dataset loading, model setup, LoRA fine-tuning, testing, and project backup steps
README.md	Dataset documentation

Dataset Format

The training examples use a messages structure with user and assistant roles.

Example:

{
  "messages": [
    {
      "role": "user",
      "content": "What is my current degree?"
    },
    {
      "role": "assistant",
      "content": "I am pursuing an M.Sc. in Data Science at Dayananda Sagar University, Bengaluru."
    }
  ]
}

Dataset Topics

The examples cover information such as:

• Current degree and university
• Programming languages and technical skills
• Major academic project
• Backend technologies
• Freelance experience
• Client projects
• DeptSync college automation project
• Database technologies
• Career interests

Model and Training

The accompanying notebook uses:

• Base model: Qwen/Qwen2.5-0.5B-Instruct
• Fine-tuning method: LoRA (Low-Rank Adaptation)
• Frameworks: Transformers, Datasets, PEFT, TRL, Accelerate
• Environment: Google Colab with GPU
• Training epochs: 3
• Per-device batch size: 1
• Gradient accumulation steps: 4
• Learning rate: 2e-4

The notebook successfully completed the training run and saved the LoRA adapter and tokenizer.

Training Result

The training run completed successfully with:

• Global steps: 9
• Epochs: 3
• Training loss: approximately 3.13
• Training runtime: approximately 13.37 seconds

These values describe the training run and should not be interpreted as a general accuracy measurement.

Testing

The notebook includes inference testing with questions such as:

• What is my current degree?
• What is my major academic project?
• What are my programming languages?
• What client projects have I worked on?

A system prompt containing the personal career information was also used during one test to provide explicit context to the assistant.

Important Limitation

This is a small personal dataset and is intended as a learning and prototype project.

The successful contextual test does not by itself prove that the fine-tuned model has memorized all personal facts independently. More examples and systematic evaluation would be required for a more reliable personal assistant.

Intended Use

This dataset is intended for:

• Learning about supervised fine-tuning
• Experimenting with LoRA
• Building a personal career assistant prototype
• Demonstrating an end-to-end Hugging Face fine-tuning workflow

Responsible Use and Privacy

The dataset contains personal career and educational information. Users should review the contents before sharing, redistributing, or using the dataset in other applications.

Do not add passwords, authentication tokens, private documents, financial information, or other sensitive personal information to the dataset.

Reproducibility

The training workflow is documented in:

personal_career_assistant_training.ipynb

The notebook can be opened and executed in Google Colab after installing the required Python packages and providing access to the dataset.

Author

C M Karthik

M.Sc. Data Science
Dayananda Sagar University, Bengaluru, India