Data Foundations Lab

As a hypothetical junior analyst for a property consultancy, learners answer a simple client question: what did New York property really sell for last year? The sales data comes from public NYC records in CSV, Excel and JSON files, each with its own problems, from encoding errors to multi-sheet workbooks. Learners load every file, manage memory on a large extract and turn prices stored as text into numbers they can trust. The project closes by running the same workflow on property sales from the United Kingdom.

A city wants to publish a daily air quality figure, but first someone has to confirm the sensor feed can be trusted. Using a year of hourly readings from a roadside sensor in an Italian city, learners find and map gaps in the data, test simple ways to fill them and learn when each method gives a misleading picture. They also protect the feed with duplicate checks, filters and range checks. The final lesson applies the same checks to air quality data from multiple stations in Beijing.

When US interest rates rose quickly in 2022 and 2023, which parts of the economy held up best? Learners answer this question using US industry returns from the Kenneth French Data Library and interest rate data from FRED. No finance background is needed, as every market concept is explained in plain language. Learners build reusable functions, create new measures such as the growth of one dollar invested, and combine data series that follow different calendars. The project closes by asking the same question of European markets.

A manager wants to understand two years of online sales and needs to trust what the charts show. Learners work with the transaction history of a UK online gift shop, where cancelled orders, missing customer IDs and repeated entries are part of the data. They build clear charts a non-technical reader can follow and record every cleaning decision in a short data quality report. The project ends by answering the same questions with purchase data from a different online shop.

A home valuation firm prices houses from county records, but many fields are empty. Some values are truly unknown, while others simply mean a house has no pool, garage or basement. Learners fill gaps using neighborhood context and rules that follow the meaning of the data, then measure how much each choice changes the result. They package their cleaning steps as reusable, tested code. The project closes by running the same pipeline on property records from Washington, DC.

A research lab placed 54 wireless sensors on one floor to track temperature, humidity and light, and the building team wants to use them to run the heating. When batteries run low, sensors start reporting impossible values. Learners design a multi-level filter, handle edge cases and test how much the result shifts when thresholds change. They then move their filter into reusable modules with automated tests. The final lesson runs the same module on weather stations along Chicago's beaches.

A personal finance newsletter wants a simple health check of every US-listed company: how much it sells, how much it keeps as profit and how much it owes. Using annual reports filed with the SEC, learners build meaningful measures such as profit margins and test how choices like trimming extreme values change the picture. No finance background is needed. Along the way, they learn to recognize why data goes missing and choose the right strategy for each case. The project closes by rebuilding the table from the SEC's raw quarterly files.

A city transport office wants to plan new bike docks using trip records from London's public bike scheme, but first it needs a dataset it can trust and a quality score it can quote. The records include rides lasting seconds or weeks, mismatched timestamps and station codes missing their leading zeros. Learners build a framework of validation rules that scores the data's reliability and produce an analysis-ready file with every transformation documented. In the final lesson, the same pipeline and rules run on bike trips from Helsinki.