DevRel AI Assistant

Fullstack AI program that reads GitHub issues and crafts real DevRel strategies — not just labels, but actions.

Comment

Inspiration

We wanted to solve a real, recurring frustration in the developer ecosystem: OSS maintainers and Developer Relations teams are overwhelmed by GitHub issues. Most feedback gets buried, community signals are hard to track, and teams spend hours manually prioritizing bugs, questions, and feature requests.

As student developers passionate about AI, we saw this as the perfect opportunity to blend open-source LLMs with a real-world bottleneck — giving rise to the idea of an AI DevRel Agent that thinks, reads, and suggests like a human DevRel Manager.

What it does

DevRel AI Assistant is a fully autonomous AI-powered GitHub analysis dashboard. You simply enter any public GitHub repository, and the system:

  • 🧠 Classifies issues using a local open-source LLM (TinyLLaMA via Ollama)
  • 🌐 Gathers real-time web context using the Tavily search API
  • 💡 Suggests clear, actionable DevRel strategies (in under 120 words)
  • 📊 Renders an interactive dashboard with charts, filters, and download options

It works offline-first, is deployable anywhere, and respects open-source AI principles — while delivering real business and community value.

How we built it

We used Python and Streamlit for the frontend interface.
The backend pipeline includes:

  • A classifier agent powered by Ollama’s TinyLLaMA running locally
  • A DevRel action agent for writing crisp, business-grade responses
  • GitHub API wrappers to fetch and clean issue data
  • The Tavily API to enrich issues with real-time external context
  • Plotly and pandas for rendering responsive visual analytics

The app is containerized using Docker and deployed on Hugging Face Spaces for free, public access.

Challenges we ran into

  • Running an LLM model on limited GPU/RAM inside Hugging Face Spaces was extremely tricky — we optimized response time by switching from larger models like Mistral to TinyLLaMA.
  • Tavily has a credit limit, so we designed a fallback mechanism for when API limits are hit.
  • Creating concise but meaningful suggestions under 120 words required several prompt engineering iterations to avoid repetition and vagueness.

Accomplishments that we're proud of

  • Fully working agent pipeline with zero cloud compute dependencies
  • High-quality DevRel suggestions generated from local LLMs
  • A polished, theme-aware dashboard with live filtering and download support
  • Successfully deployed on a free-tier public domain with full functionality

What we learned

  • How to optimize LLM behavior through prompt tuning and task splitting
  • Real-world application of LLMs beyond just chat interfaces
  • How to serve AI tools at scale using Streamlit + Hugging Face
  • The power of combining agentic workflows, open APIs, and local models

What's next for DevRel AI Assistant

  • Add support for multi-repo trend analysis and historical insights
  • Include voice input for accessibility and hands-free feedback
  • Extend to Bitbucket and GitLab support
  • Enable Slack and Discord bot integrations for real-time community updates

We're also planning to open-source this as an installable DevRel CLI/Agent for teams wanting private GitHub insights powered by their own local models.

Built With

Share this project:

Updates

Submission history