Machine Learning Specialist

Adeptiv AI is an enterprise AI Governance platform purpose-built for regulated industries Banking, Financial Services, Insurance, and Healthcare.

We are hiring a Machine Learning Engineer specialised in ML Bias, Fairness, and Explainability. Candidate must have experience with tools such as IBM AIF360, AWS SageMaker Clarify, Fiddler AI, Arthur AI, SHAP, LIME, etc.

Important Note: This is NOT a Data Science or ML Modelling role. You will NOT be training models. But this is more of an ML Governance role for ML Models, Bias, Fairness, and Explainability. So please apply only if you have experience in this.

Experience:

- Minimum 5+ years of relevant experience in Python, Machine Learning, ML Models, Bias & Fairness evaluation, etc.

- Deep, working knowledge of group fairness metrics: Demographic Parity, Equalized Odds, Equal Opportunity, Disparate Impact not just names, but mathematical definitions and when each is appropriate.

- Practical experience with IBM AIF360: working with ClassificationMetric, BinaryLabelDataset, and the pre/in/post-processing mitigation algorithms at the code level

- Understanding of SHAP and LIME for model explainability global feature importance, local per-decision explanations, counterfactual explanations

- Experience with regulated industry AI use cases, such as credit scoring, fraud detection, insurance underwriting, clinical decision support is a huge plus

Responsibilities:

- AIF360 adapter parse ClassificationMetric JSON output; map AIF360 metric names to our canonical metric registry

- SageMaker Clarify adapter parse bias_report.json from S3; handle both pre- and post-training metric sections

- Fiddler AI adapter API-driven fetch of fairness monitoring results via Fiddler REST client

- Arthur AI adapter integrate with Arthur's evaluation and monitoring API

- PDF/report adapter use LLM-assisted extraction to parse unstructured model validation reports; fuzzy-map metric names to canonical IDs; flag for human review

- CSV / JSON manual upload template-driven ingest with column validation

- Group fairness: Demographic Parity Difference, Equalized Odds Difference, Equal Opportunity Difference, Disparate Impact Ratio, Average Odds Difference, Theil Index

- Individual fairness: Consistency Score, Counterfactual fairness measures

- MLflow or similar MLOps platforms reading model metadata, performance logs, connecting model versions to evaluation records.