Hi 
community,
I’m excited to share an applied NLP project I recently open-sourced:
GitHub: AI-Powered Ticket Routing & SLA Breach Prediction in JIRA
What It Does:
This tool uses transformer-based models (fine-tuned on internal support data) to:
- Predict SLA breach probability for incoming tickets.
- Automatically classify and route tickets based on historical handling patterns.
- Generate alerts for high-risk cases using a trained classifier.
It’s designed for real-time integration into JIRA environments, especially helpful for large support teams under SLA pressure.
Tech Stack:
- Hugging Face Transformers (
bert-base-uncased) - Scikit-learn for baseline models
- Pandas, FastAPI, JIRA REST API
- Model interpretability with SHAP
What I’m Looking For:
I’d love to get feedback from the community on:
- Better NLP architectures for short, noisy, multilingual support tickets
- Suggestions for incorporating zero-shot classification (e.g.,
facebook/bart-large-mnli) - Ideas for improving label imbalance in ticket outcomes
- Ways to optimize model latency for real-time usage in MLOps pipelines
Bonus:
If you’ve worked on similar use cases in ITSM, support automation, or task classification, I’d love to connect and hear your thoughts!
Thanks for checking it out — and feel free to 
the repo if you find it useful!
Looking forward to your insights 
— Arooj Javed