Learning Tracks
Grouped by domain, one directory per topic. Each topic is a self-contained track driven by the teach skill: a mission, an ordered set of lessons, and the reference material those lessons earned.
Topics
| Domain | Topic | Mission | Lessons |
|---|---|---|---|
| programming | Bash | Write shell that survives production: quoting, exit status, the failure modes, and knowing when to stop and use a real language | 17 |
| programming | C# | Own a C# service: the type system, async, and what the CLR does with what you wrote | 45 |
| programming | Go | Own Go on a team: design, ship and operate a production service | 41 |
| programming | Java | Own a Java service: model it in modern Java, then operate what the JVM does with it | 52 |
| programming | Kotlin | Own Kotlin on the JVM: null safety, coroutines, and where the Java habit produces Kotlin that compiles and reads badly | 44 |
| programming | Python | Own Python in production: ship it typed, tested, packaged and profiled | 44 |
| programming | Rust | Own Rust: design with ownership instead of fighting the borrow checker, then ship the crate | 67 |
| programming | SQL | Own the database: write the query, read the plan, design the schema, survive the concurrency | 53 |
| programming | TypeScript | Own a TypeScript codebase: make the compiler reject the states that should not exist | 49 |
| llm | Adapter fine-tuning | Decide whether to fine-tune, run it, prove it worked, ship it | 27 |
| llm | Pretraining | Train a model from scratch at scale: data, tokenizer, distributed training, and a defensible compute budget | 24 |
| llm | Post-training | Align a base model: SFT, reward modeling, RLHF, DPO, and GRPO, and knowing which one to pick | 20 |
| llm | Inference | Serve a model: the KV cache, batching, quantization at serve time, and a latency budget you can defend | 24 |
| llm | Evals | Prove a model change helped: build the eval, hold out the data honestly, and defend the number against contamination | 18 |
| llm | RAG | Own retrieval: chunking, embeddings, hybrid search, reranking, and measuring whether the right thing was retrieved | 19 |
| llm | Transformers | Build attention from scratch, so the architecture stops being a black box behind the training script | 22 |
| llm | Agents | Build the loop yourself: tools, context, memory, and knowing when an agent is the wrong answer | 37 |
| data | Postgres | Operate Postgres: the WAL, vacuum and bloat, replication, and what an index costs to keep | 18 |
| data | Kafka | Own the log: partitions, consumer groups, delivery guarantees, and what "exactly once" actually costs | 16 |
| data | Redis | Use Redis for what it is: the memory model, eviction, persistence, and the patterns that quietly misuse it | 17 |
| architecture | Distributed Systems | Reason about partial failure: what a network can do to you, which consistency you are actually buying, and why consensus is expensive | 19 |
| architecture | API Design | Design an interface others depend on: the contract, its errors, and how it changes without breaking them | 19 |
Starting a topic
cp -r templates/learning-workspace learning/<domain>/<topic-slug>
Then run the teach skill and name the topic. It reads the track, fills in the mission by interviewing you, and writes the first lesson.
A domain is a grouping, not a track: it holds topic directories and nothing else. Add a new one when a second topic would share it.
Layout
learning/
├── <domain>/ # programming, llm, …
│ └── <topic>/ # one track per topic
├── .nav.yml # sidebar order/titles for the published site
└── README.md # this index
Each track:
learning/<domain>/<topic>/
├── README.md # mission, success criteria, lesson index
├── GLOSSARY.md # canonical terms for this topic
├── RESOURCES.md # trusted sources, known gaps
├── NOTES.md # preferences and working notes
├── lessons/ # NNNN-slug.md, one tight win each
├── reference/ # cheat sheets, built for lookup
└── learning-records/ # NNNN-slug.md, what was demonstrably learned
Lessons number from 0001 within each track. Tracks may link to each other, but a link is a pointer rather than an inclusion: it does not pull another topic's material into this mission. Prefer linking a glossary term or a reference sheet over a lesson, because those are written for lookup and land on any reader.
The teach skill owns the format of every file here. Read its spec before hand-authoring one.
License
Everything on this site is CC BY 4.0. Copy it, translate it, teach from it, build on it, commercially or not. Just say where it came from.