Financial institutions regularly estimate possible portfolio losses arising from (future) changes in the market values of their assets, e.g. using the Value at Risk (VaR) metric. We show a VaR model using Filtered Historical Simulation and illustrate some of the usual processes for calibrating, validating and analysing these models, such as calibration and backtesting. Although the model is relatively simple, the challenges posed by their different requirements - flexibility, data volume, performance - inevitably lead to many shortcuts and complexity. With the ultimate goal of managing this complexity, we explore and evaluate, with concrete coded examples, the PyData (e.g. Pandas, Bokeh) and Apache Spark frameworks from a practitioner's point of view.