ML Engineering

Inside the work.

A hands-on companion to the research collection: architecture, tools, adaptations, code and the checks that keep an implementation honest.

Why 61 articles?

The starting index selected 61 evidence files. They include code, saved results, operational notes and an upstream package. They are not 61 independent models or 61 new scientific contributions. Result articles share an architecture with their implementation companion and focus on a different engineering question.

What is original, and what is upstream?

The commentary and diagrams are newly authored from the inspected archive. Short excerpts retain their original text and hashes. TensorNEAT is explicitly attributed to its upstream authors; PyTorch, Keras, JAX, Flax, EvoJAX, SAC, SIREN and other established tools are not presented as inventions of this project. Presence in a personal archive is not independent proof of authorship or redistribution rights. Public release requires the owner’s rights review, including any adapted third-party code.

Historical code, current explanation

We do not silently repair excerpts. Implementation limitations and proposed checks are explained beside them. Historical environments are not fully reconstructed, and the snippets are not advertised as standalone runnable applications. Official documentation provides API context; it does not prove the archive used the current library version.

Evidence and privacy

The private source index is an input to a read-only audit, not a site asset. No interview prose, credentials, raw recordings, full operational runbooks or model weights are included. Paths are relative. Only allowed subdirectories were inspected; files directly under the engineering root and RuView were excluded.

What was executed?

The build, source-integrity tests, static-site checks and small synthetic mechanism checks run locally. No archived training jobs, device acquisition, broker connections, hyperparameter search or production changes were executed. Suggested article-specific follow-up checks are labeled as suggestions.

Two branches, one collection

Explore ML Engineering · Research stories · Scientific manuscripts · All 61 articles