GitMarvel

Visualize, Analyze, and Level Up Your Code on GitHub

Comment

Inspiration

The idea for GitMarvel arose from our desire to make coding more interactive and insightful for developers. While GitHub provides powerful tools for collaboration, we wanted to enhance its capabilities by using SambaNova’s AI models with storytelling and analysis. Our goal was to create a platform that acts as a developer's companion—providing valuable insights, actionable feedback, and an engaging way to track progress.

What It Does

GitMarvel is a comprehensive AI-powered web app that transforms the GitHub experience by introducing several key features:

  • Code Reviews: Uses Meta-Llama-3.1-8B-Instruct to analyze repository context and provide specific, actionable feedback on a given commit hash.
  • Code Storytelling: Generates a cinematic narrative of a repository’s progress using chronological data analysis, enhanced with images fetched from the Pixabay API.
  • Profile Analysis: Performs SWOT analysis on GitHub profiles, identifying strengths, weaknesses, and growth opportunities.
  • README Generator: Leverages Meta-Llama-3.2-3B-Instruct to analyze repository files and create structured, professional README files.
  • Dynamic Chat: Utilizes a pre-filter API call to determine the most suitable model for user queries, selecting from models like Meta-Llama-3.2-1B-Instruct, Meta-Llama-3.2-11B-Vision-Instruct, and others. The chat provides context-aware answers on topics like repositories, code, or general programming queries.
  • Reports: Exports profile and code review analyses as downloadable HTML reports for offline use.

How We Built It

  1. Backend: Built with Flask to process GitHub API requests, handle model interactions, and generate insights.
  2. AI Models:
    • Used Meta-Llama-3.1-8B-Instruct for most tasks, including code reviews and general conversation.
    • Incorporated dynamic model selection for chat, with the pre-filter API deciding between:
      • Meta-Llama-3.2-1B-Instruct
      • Meta-Llama-3.2-3B-Instruct
      • Meta-Llama-3.2-11B-Vision-Instruct
      • Meta-Llama-3.1-405B-Instruct
    • Fine-tuned prompts for optimal model responses across diverse tasks.
  3. Frontend: A HTML-CSS based UI deployed on Vercel, designed for simplicity and user intuitiveness.
  4. APIs Used:
    • GitHub API: For fetching repository details, commit histories, and profile data.
    • Pixabay API: To generate visuals for storytelling features.
    • Sambanova API: For our core features.

GitMarvel Architecture

Challenges We Ran Into

  • Managing dynamic model selection to ensure accurate responses across diverse queries.
  • Optimizing the application to handle large commit files efficiently while maintaining performance.
  • Implementing multiple features within a constrained timeline while ensuring seamless user experience.

Accomplishments That We’re Proud Of

  • Successfully integrated SambaNova’s models into our web application.
  • Built a dynamic conversational AI system that adapts responses using a model selection pipeline.
  • Designed and deployed GitMarvel on Vercel as a polished, user-friendly application.

What We Learned

  • Developed expertise in integrating and managing large AI model pipelines dynamically.
  • Strengthened backend development skills, particularly with Flask and multi-API handling.
  • Gained valuable insights into user experience design for technical tools.

What’s Next for GitMarvel

  • Enterprise Integration: Add support for large organizational repositories and team-wide analytics.
  • Historical Data: Store user interactions and previous insights for enhanced personalization.
  • Public Launch: Scale infrastructure to handle higher user loads and improve app stability.

Built With

Share this project:

Updates