I research music through embeddings and narrative arcs, and study how AI systems interpret meaning and remember evidence. Close reading, listening, and computational experiments help me examine how ideas develop across songs and albums. I also help engineering teams evaluate their own systems and decide what to change. Choose the path that fits what you want to do.

Understand a question

What must an agent remember to explain an album? Start with What Should an Agent Remember?. It compares ways of retaining evidence across four albums and explains why a fluent summary can still lose the relationship a question asks about. You can follow the argument without running code.

Then read Memory Is Not Context: a smaller final context can conceal many additional model calls, and an automated verifier can accept an unsupported interpretation.

For the musical questions behind this work, explore Narrative Arcs in Music.

Inspect an experiment

The experiment index links the study designs, code, recorded outputs, and review materials. Each entry says whether it is a measured study, an exploratory analysis, or an educational walkthrough.

Begin with the album-memory experiments. You can inspect published results without making paid API calls. Repeating live model runs is a separate step with its own requirements and costs.

Read the methods and corrections notes alongside the results. Model-generated judgments are not independent human validation, and a result on selected albums is not a finding about every artist or application.

Apply the method

For a team operating an AI application, start with Prompts Are Release Artifacts. It connects observed failures, versioned prompts, evaluation, and release decisions using a synthetic support-documentation example.

If your team needs to investigate retrieval, memory, or unsupported answers, read about the AI reliability and evaluation diagnostic, with options for architecture review, optimization, and ongoing technical guidance.

Explore the wider archive

  • AI Engineering — evaluation, prompts, infrastructure, and production systems.
  • Music Research — narrative, multilingual lyrics, and computational interpretation.
  • Edge + Agentic AI — inference and constraints on smaller devices.
  • All writing — the complete chronological archive.

Receive new essays by email or follow the RSS feed.