I will audit your ml model
About this Gig
DOES YOUR ML MODEL ACTUALLY DO WHAT YOU THINK IT DOES?
Most models go live with an accuracy score and a prayer.
Accuracy alone is misleading a model that labels everything 'not fraud'
gets 95% accuracy but catches zero fraud. Your CTO, Head of Risk, and
Compliance team deserve better than that.
I will give them that.
WHAT YOU GET (all 5 phases of a professional audit)
Phase 1 Discovery & Data Quality
Phase 2 Performance Analysis
Phase 3 Explainability (SHAP)
Phase 4 Fairness & Bias (Fairlearn)
Phase 5 Reporting
Programming language:
Python
•
Colab
•
MLflow
Frameworks:
Scikit-learn
•
DeepPy
•
Keras
•
PyTorch
•
Panda
Tools:
Jupyter Notebook
•
MLflow
FAQ
My model isn't scikit-learn. Can you still audit it?
Yes. I can audit any classifier that exposes predict() and predict_proba() methods, including XGBoost, LightGBM, CatBoost, PyTorch, and TensorFlow. For models behind an API endpoint, I use your API to generate predictions. The SHAP explainability step uses TreeExplainer for tree models and KernelExp
I can't share real customer data. Can we still work together?
Absolutely. I work with anonymised or synthetic datasets regularly. I can also work with a representative sample (minimum 500 rows), or we can sign an NDA and use a secure VPN/API access arrangement. Just message me before ordering and we can discuss the right approach.

