Linear regression is a core component of many useful ML tasks. Here is a first attempt at an embedded-optimized implementation, using gradient descent with support for L1+L2 regularization (often called ElasticNet).
Got some assistance from Claude with this code, though it took some heavy-handed guidance and frequent whipping with the core C code. It however did super well in exposing this as a MicroPython module (I gave an existing module as reference), and was very useful in creating some quick but realistic tests.