Myles MellorCommercial, digital & marketing operator
← Systems

Making podcasts and courses compound, not evaporate

A learning system that turns passive listening, reading and courses into structured records — separating what I learned from the ideas it sparked, and scoring the result against real evidence.

Most of what you take in from a podcast or article is gone within a day. I built a pipeline that turns each one into a structured record on a single schema, keeps 'what I learned' apart from 'the ideas it sparked', synthesises across them, and feeds an evidence-scored skills matrix — so exposure compounds into capability I can actually point to.

I listen to a lot of podcasts and read a lot about this field. Left alone, almost none of it sticks — you finish an episode with a vague sense of "that was useful" and by the next day it's gone. The usual fixes (highlights, a notes app, a tag dump) don't help, because the problem isn't storage. It's that raw exposure doesn't compound. So I built a small system to make it.

From a podcast to a structured record

Every input — a podcast, an article, a course — gets turned into a structured record on one canonical schema: the same shape every time, capturing what was claimed, what's worth keeping, and how it connects to what I'm building. Because the records share a schema, they're comparable and searchable rather than a pile of free-text notes — and they're indexed, so I can ask the whole collection questions later instead of trying to remember which episode said what.

The discipline that makes this work is not trying to keep everything. Most of any given episode is noise; the record captures the few things that earned their place. Capture is generous; keeping is selective.

Two streams, kept apart

The non-obvious design decision is splitting each record into two separate streams:

  • Knowledge — what I actually learned. Facts, techniques, how something works.
  • Ideas — what the episode sparked. Things I might build or try.

These look similar and are completely different. Conflating them is a classic failure: an exciting idea gets filed as if it were established knowledge, or a solid fact gets lost among half-formed maybes. Keeping them apart means the knowledge stays trustworthy and the ideas stay visible as ideas — captured, not yet committed to. (It's the same instinct that runs through the rest of the system: capturing something creates a memory, not a commitment.)

Compounding: synthesis and measurement

Records on their own are just a tidy archive. Two things make them compound:

  • Synthesis. Periodically I pull a span of episodes into a bundle — a synthesis across many inputs that surfaces the patterns no single episode showed. The value isn't in any one record; it's in what they say together.
  • Measurement, scored against evidence. All of this feeds a skills matrix — a map of my actual AI capability by area. The part that keeps it honest: each level is scored against real evidence (actual files, commits, shipped work), checked by a periodic scan that re-verifies the claims and flags any the evidence no longer supports. Gaps are shown, not hidden. It's the opposite of a self-assessed skills list — the matrix is only allowed to claim what the work proves. (That's what the capability map is built from.)

Why bother

The honest reason is that capability built this way is demonstrable. Anyone can say they "keep up with AI." Far fewer can show a structured trail of what they took in, what they did with it, and a capability claim that's pinned to evidence rather than to confidence. The system turns a habit that usually produces nothing durable into one that produces a compounding, inspectable record — a trail I can point to, not a claim you have to take on trust.