DuckDB: query a client's CSV and Parquet files on your laptop before you ask for infrastructure
An FDE Times tools guide walks through embedded DuckDB for day-one client data checks—read_csv sniffing pitfalls, Parquet column pruning, and memory limits when converting large CSVs on a constrained laptop.
FDE Times published a practical guide aimed at Forward Deployed Engineers who receive a folder of client CSV and Parquet exports and need answers before the next meeting—without waiting on a cluster. DuckDB runs embedded in-process under an MIT licence, so there is no separate server to install on a client laptop.
The piece stresses checking what read_csv guessed for column types with DESCRIBE, because the sniffer only samples by default; wrong date types corrupt the numbers you put on slides. For Parquet it explains column pruning and pushdown filters, and for larger-than-RAM CSVs it covers converting once with COPY, setting memory_limit, and comparing row counts before trusting the new file.
For FDEs, the takeaway is operational: before requesting infrastructure, open DuckDB, validate types, answer the client's question in SQL, and only then design a pipeline—skills that show up clearly on a CV when tied to messy client files and same-day answers.
Source: fdetimes.net