Latest Insights
Exploring the frontiers of AI, software architecture, and engineering.
The Floating-Point Problem: Why Your Sums Are Wrong
0.1 + 0.2 is not 0.3 in any language. Floating-point arithmetic drifts, cancellation destroys precision, and your grand totals quietly inherit the damage. Here is how it works and what to do.
Read Article →Measuring Your Forecast: MAPE, RMSE, and the Holdout
A forecast is only as good as its error metric — and every popular error metric lies in a different way. Here is how to evaluate forecasts with a holdout and pick the right measure.
Read Article →Missing Data: When to Drop, When to Impute, When to Worry
Missing values are not one problem — they are three, with different remedies. The pattern of missingness decides whether you can drop, impute, or must investigate before doing anything.
Read Article →Anomalies in Time Series: Spikes, Level Shifts, and Seasonal Surprises
A spike is not a level shift, and neither is a seasonal dip. Anomaly detection on time series fails when it treats all three the same. Here is the residual-based approach that tells them apart.
Read Article →Parsing Megabytes Without Freezing the UI
A 50 MB CSV should not freeze a browser tab. Chunked parsing, Web Workers and transferable buffers keep the interface alive — here is the engineering pattern.
Read Article →Shipping Your Analysis: Choosing the Right Export Format
Every export format lies in a different way. CSV loses types, JSON loses labels, HTML keeps everything human-readable. Match the format to the consumer and the decision.
Read Article →