SubinoyBera/Toxic-TweetTagger
End to End Hate Tweet Detection - a production ready ML application!
What's novel
End to End Hate Tweet Detection - a production ready ML application!
Code Analysis
19 files read · 4 roundsAn end-to-end hate speech detection system using XGBoost with TF-IDF features, wrapped in a production-oriented MLOps pipeline (DVC, MLflow, Evidently) with a FastAPI inference service, LIME explainability, Prometheus monitoring, and async MongoDB logging.
Strengths
Well-structured production architecture with clean separation of pipeline stages, async batch MongoDB writes, Prometheus observability, confidence-margin warnings, and LIME explainability. The code is consistently organized with typed config entities, custom exceptions, and proper dependency injection in the API layer.
Weaknesses
The core ML is a standard XGBoost+TF-IDF classifier with no novel algorithmic contribution. There are notable bugs (inverted validation logic in data_validation.py, module-level EnvironmentError in model_evaluation.py, trivially-true test assertions) and the API design pushes preprocessing to the client which is unusual and error-prone.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
153
Contributors
1
Files
196
Active weeks
11
Repository
Language
Python
Stars
1
Forks
0
License
Apache-2.0