Cognoria

Why we exist. There’s a skills gap: too many projects ship without fundamentals. We’re here to fix that through disciplined practice.

How we’re different. Small, selective, and built around doing and teaching one another - no lectures, no content treadmill.

Outcome. Join labs, lead in your domain, or build - this is a launchpad, not a chatroom.

Who this is for

Engineers/analysts who want to understand and implement language models from first principles, ship code every week, and be challenged by peers.

Who this isn’t for

Passive listeners, “prompt-only” builders, or anyone unable to commit 8–10 hours weekly. This is not a casual reading group.

Why join (briefly)

• Mastery by doing. Every week ends with code, tests, learning-curve plots, and a short reflection.
• High-signal peers. Small pods, direct feedback, everyone contributes.
• First principles → real systems. Fundamentals you’ll actually use, paired with language-model work.
• Career credibility. Public repos and clear write-ups that prove you can build.
• No fluff. Curated canon, zero content treadmill.

What we’ll do (curriculum evolves with the cohort)

Two lenses that meet in the middle:

• Bottom-up (Fundamentals): vectorized loss/gradients, backprop, initialization/regularization, optimization and numerics, clean PyTorch practice.
• Top-down (Language Modeling): tokenization, attention, decoder blocks, training loops, decoding, and small evaluation; later, practical finetuning workflows.

We’ll publish a living canon to members; sources may change as we learn.

How we work (rituals, not lectures)

• Rotating facilitation: a different member guides each session.
• Challenge is everyone’s duty: ask “show me in code,” stress-test assumptions, be kind and direct.
• Short discussion, then build.
• One helpful review most weeks on a teammate’s contribution (usually a PR or short note).
• Periodic check-ins and demo moments.
• Sessions recorded for members. No observers.

What joining entails (the bar, briefly)

A weekly contribution (usually a PR or short note), a helpful review most weeks, active participation in a pod, and basic norms: be decent, credit work, don’t plagiarize.

Founding cohort

Small and selective. We’ll set the schedule with the cohort.
We are a learning pod, not a product.

Values

First principles > shortcuts · Show me in code · Challenge is a duty · High signal, low noise