jeff-calderon/imaginator-skyfall

Dataset Card for GitHub Profiles for Resume Generation

Dataset Summary

This dataset contains curated GitHub user profiles designed to fine-tune language models for the task of professional resume writing. It is composed of profiles that meet a "high-quality" standard based on technical complexity, rich documentation, and community engagement (the "Three C's"). The goal is to train models that can generate compelling, evidence-based resume content from a developer's raw bio and project history.

Supported Tasks & Leaderboards

• text-generation: This dataset is primarily intended for fine-tuning models for text generation, specifically for creating resume summaries and project descriptions.
• summarization: The data can also be used for summarization tasks, where the model learns to condense a developer's career and projects into a concise, impactful narrative.

Languages

The dataset primarily consists of English-language content from GitHub profiles. The code repositories associated with these profiles are predominantly in the following languages:

• Python
• C++
• Go
• Rust
• TypeScript
• JavaScript

Dataset Structure

The dataset is provided in a single JSONL file (github_profiles_prepared.jsonl). Each line in the file is a JSON object representing one training example. The structure is formatted for conversational fine-tuning (e.g., with OpenAI models):

{
  "messages": [
    {
      "role": "system",
      "content": "You are a professional resume writer. Rewrite the following developer bio and project list into a compelling, evidence-based resume."
    },
    {
      "role": "user",
      "content": "CANDIDATE BIO:\\n{bio}\\n\\nKEY OPEN SOURCE PROJECTS:\\n{formatted project list}"
    }
  ]
}

Data Fields

• messages: An array of message objects.
  • role: The role of the speaker, either system or user.
  • content: The text of the message.

Data Splits

Currently, the dataset consists of a single split: train. No validation or test splits are provided at this time.

Data Collection Process

Source Data

The data was collected from public user profiles on GitHub.

Data Collection

The data was collected in a two-stage process:

1. Harvesting: A Python script (scripts/harvest_github_profiles.py) was used to search for GitHub users with a high follower count and repositories in target languages.
2. Filtering & Preparation: The harvested profiles were then filtered based on the "Three C's":
   • Complexity: At least 5 public repos in "hard" engineering languages.
   • Context: A bio of at least 10 characters and the presence of READMEs in their top repositories.
   • Clout: At least one repository with more than 50 stars.

A second script (scripts/prepare_for_llm.py) then formatted the qualifying profiles into the final JSONL structure.

Additional Information

Licensing Information

The dataset itself is released under the MIT License. However, the underlying content (user bios, repository information, and READMEs) is subject to the terms of service of GitHub and the licenses of the individual repositories.

Citation Information

@dataset{
  author={aggressor-FZX},
  title={GitHub Profiles for Resume Generation},
  year={2025},
  url={https://huggingface.co/datasets/jeff-calderon/ResumeData}
}

Contributions

Contributions are welcome! If you have suggestions for improving the dataset or the collection process, please open an issue or pull request in the GitHub repository.