Skills in practice.

AWS

• Installed CloudWatch monitoring on trading VPSs
• Built CloudWatch dashboards for RAM and storage usage
• Configured variable CloudWatch sampling rates
• Set up AWS billing alarms
• Configured AWS SQS messaging between separate VPSs
• Created restricted AWS permissions for SQS access
• Created and set custom AWS permissions profiles

• Built an automated options trading bot with IBKR
• Aligned IBKR trading bots with back-testing software
• Experience with TCP and REST API integration
• Created modular multi-threaded data processing layers
• Refined limit-order logic using live fill data to maximise premium
• Refined execution logic using observed live fills
• Prevented duplicate bot instances using unique resource keys
• Reconciled internal bot state against broker positions and orders
• Logged multi-threaded trading activity to custom text files
• Configured cloud infrastructure for automated trading systems

Back-testing

• Used QuantConnect for back-testing
• Used WhisperTrades for back-testing
• Tested EMA crossover strategy variations
• Tested Donchian breakout strategy variations
• Tested Bollinger Bands strategy variations
• Tested machine learning strategies
• Evaluated the statistical significance of back-testing results
• Experimented with back-testing the variance risk premium on SPX
• Analysed ES futures volatility using machine learning
• Modelled slippage and trading costs within back-testing

Machine Learning

• Built Explainable Boosting Machine volatility models
• Predicted final-hour realised volatility from intraday data
• Used realised volatility as a model baseline
• Built additive corrections to baseline volatility estimates
• Used Gamma regression with a log link
• Tuned EBM learning and regularisation parameters
• Used early stopping during model training
• Tested model stability across multiple random seeds
• Compared predictions across independently trained models
• Used rolling training windows for model evaluation
• Built chronological walk-forward validation pipelines
• Prevented look-ahead bias in model evaluation
• Tested quarterly model retraining schedules
• Applied decay weighting to historical training data
• Tested different training-data decay rates
• Applied monotonic constraints to selected features
• Integrated ML predictions into trading back-testing
• Analysed model robustness across different market periods
• Built models around incomplete historical feature availability
• Tested sensitivity to training-window choices
• Used SQL and Python to prepare training data from local files