Senior Machine Learning Engineer — Custom Model Development

We need a senior ML engineer to design and train a production-quality model from messy, real-world data. This is not a notebook demo. We need a well-reasoned model, a rigorous evaluation setup, and a clear record of what was tried and why. The problem We have a labeled dataset with class imbalance, noisy labels, and mixed feature types (structured fields plus free text). The goal is a model that generalizes on unseen data, not one that only looks good on a random split. What you’ll do Audit the dataset: leakage, label quality, missingness, outliers, and train/validation/test design Build a reproducible preprocessing and feature pipeline (structured features + text embeddings if useful) Train and compare several approaches (e.g. gradient boosting, linear baselines, and a neural/transformer option) Handle imbalance and noisy labels with the right sampling, loss, or calibration methods Run proper validation (stratified or grouped splits — no row-level leakage) Error analysis: where the model fails, which classes confuse, and what would improve it next Deliver trained artifacts, training/eval code, metrics, and a short technical write-up Success looks like A holdout evaluation you can defend (precision, recall, F1, PR-AUC, calibration) A baseline vs. final model comparison, not a single lucky run Code someone else can retrain without reverse-engineering your notebook What we’re looking for Senior experience building models in Python (PyTorch or TensorFlow, scikit-learn, XGBoost/LightGBM, pandas) Strong instincts on validation design, leakage, and overfitting Comfortable with both classical ML and deep learning; pick the simpler model if it wins Clear written communication and independent execution NLP / transformer experience is a plus To apply, send 1–2 relevant model-build case studies (what the data looked like, what you tried, and the final metrics) plus your rate.