Machine Learning Engineer

We're looking for a part-time Machine Learning / Quant Research Engineer to work on predictive modelling for financial market data.

The core work is already past the raw-data stage. We have a large time-series dataset with engineered market features across multiple rolling horizons, including price, order flow, liquidations, volatility and related market-state measurements. We are now building the research layer that maps these market states to future outcomes.

The role will focus on designing, implementing and validating machine-learning experiments around this dataset.

• Build regression and classification models for financial time-series data.
• Develop strong baselines using linear/ridge models, decision trees, Random Forest and related methods.
• Build and tune XGBoost/LightGBM/CatBoost models for large tabular datasets.
• Identify nonlinear interactions across features and multiple time horizons.
• Test whether new feature groups add incremental predictive information.
• Run feature-family and horizon-level ablation studies.
• Build leakage-safe chronological and walk-forward validation pipelines.
• Work with multiple prediction targets across different future horizons.
• Analyse residuals and identify areas where existing models systematically fail.
• Use tools such as SHAP, permutation importance and model diagnostics to understand learned relationships.
• Explore PCA, clustering and other unsupervised methods for dimensionality reduction, redundancy analysis and market-state/regime discovery.
• Run controlled hyperparameter searches and GPU-accelerated experiments where appropriate.
• Build reproducible research pipelines that make it easy to compare models, features and experiments.

Strong practical ability with:

• Python
• pandas / Polars / NumPy
• scikit-learn
• XGBoost, LightGBM or CatBoost
• Regression and classification
• Tree-based models and gradient boosting
• Feature engineering and feature selection
• Time-series validation
• Overfitting, regularization and model selection
• Data leakage prevention
• Model evaluation and statistical reasoning

You should understand the difference between a model fitting historical data and a relationship genuinely surviving unseen data.

Experience with financial markets is useful but not mandatory. Strong statistical and machine-learning thinking matters more than knowing trading terminology.

Experience with PyTorch, CUDA/GPU workloads, PCA, clustering, SHAP or large time-series datasets is a plus.

This is a research-heavy role. We care less about implementing complicated models for their own sake and more about running clean experiments, understanding why something works, challenging apparent results and building evidence step by step.

You'll work closely with the founder on model design and research questions, while taking ownership of implementation, experimentation and diagnostics.

This is a part-time role with the potential to expand based on fit and research progress.