README.md

metadata

title: GitHub Issue Triager
emoji: 🎫
colorFrom: yellow
colorTo: blue
sdk: gradio
app_file: app.py
pinned: false
license: mit

GitHub Issue Triager

Question

Can an LLM turn a messy issue report into useful project-management structure?

System Boundary

This Space is a triage assistant. It does not make final product decisions; it proposes labels, priority, severity, assignment hints, and next steps for human review.

Method

The issue text is passed to an instruction model with a structured JSON schema. The app parses the response and presents both the classification and the reasoning.

Technique

This is structured generation. Instead of asking the model for a paragraph, the app asks for a bounded object with fields that can be validated and used by another system.

The schema is the interface between language understanding and workflow automation.

Output

The app returns a title, labels, priority, severity, category, assignee suggestion, estimated effort, and rationale.

Why It Matters

LLMs become more useful in operations when their output is constrained into a schema. This demo shows how unstructured text can become a reviewable workflow artifact.

What To Notice

The rationale is important. A label without reasoning is hard to review; a label with reasoning can be corrected and used to improve future prompts or datasets.

Effect In Practice

Issue triage can reduce repetitive coordination work and make human review more consistent, especially in repositories with many incoming reports.

Hugging Face Extension

This can grow into a labeled issue-triage dataset and a model-comparison Space for classification accuracy, calibration, and rationale quality.

Limitations

The model does not know repository history unless the text includes it. Production triage should integrate repository metadata, issue history, code ownership, and feedback loops.

Run Locally

pip install -r requirements.txt
python app.py