BLOG / ENGINEERING NOTES
Ideas worth exploring.
Data science and AI engineering, with an eye on what matters: understanding what we are building and what it is useful for.

Evaluating the safety of an AI application
RAG, databases and tools make LLM applications useful, but they also change their risks. How to test that they protect people while still helping them.

Teaching AI to build scientific models
An equation can reproduce the data without explaining how a system works. What AI needs to move from fitting observations to scientific understanding.

What is an AI harness? From model to action
Why can two assistants using the same model handle a task so differently? The harness helps explain the difference.

A network that does not start from scratch
Word by word, a recurrent network updates its memory. Open the machine to see how it combines input with context, and what happens when that connection is removed.

Understanding the relational model: from tables to rules
Relations, tuples, attributes, domains, schema, degree and keys: an explanation from the ground up, using a small shop.

Markov chains: where a fleet ends up
Three stations, a hundred bicycles and a question about their next move. Build a Markov chain from a concrete situation.