How we teach
The same delivery and assessment model applies to every one of our 115 courses. It is designed so that what you claim you can do is backed by evidence you produced yourself.
The 35 / 20 / 45 model
Principles
Mathematical, physical and architectural foundations. You cannot optimize a cache hierarchy or a scheduler you cannot reason about, so we build the model first — from source code and hardware documentation, not folklore.
Guided investigation
Tracing, fault injection, comparison and failure analysis. You learn to interrogate a running system: ftrace and perf on the kernel, protocol analysers on the wire, counters on the silicon.
Engineering studio
Implementation, integration, testing and measurement. The largest block of every course is you building the thing, breaking it, and proving what it does.
The evidence portfolio
Every course ends with a capstone, and every capstone is judged on its evidence, not its narrative. A passing submission contains all eight of these:
- Source repository and immutable submission tag
- Toolchain, dependency, hardware/VM and configuration manifest
- Automated unit, integration and system tests
- Raw machine-readable measurements plus the plotting and analysis scripts
- Failure-injection or negative-test results
- Architecture and design record explaining the tradeoffs
- Reproduction instructions from a clean environment
- Final report separating observed facts, interpretations, assumptions and unverified claims
This is deliberately the same discipline a senior engineer is held to when landing a kernel patch or signing off a board bring-up. The portfolio is yours to keep and to show.
Assessment rubric
| Component | Weight |
|---|---|
| Principlesderivations, architecture reasoning, design decisions | 20% |
| Investigationsdiagnosis, profiling, fault injection, interpretation | 20% |
| Capstone implementationcorrectness, integration, engineering quality | 40% |
| Evidence & reproducibilitytests, measurements, manifests, report, limitations | 20% |
Lab modes
Software-only baselines (QEMU, simulators, cloud GPU) wherever that is technically honest. Where hardware changes the answer — board bring-up, timing, DMA coherency — the course states its hardware mode up front, and in-person cohorts get the real boards.
See what a lesson actually looks like
One complete lab, start to finish, at the depth every course is taught.