Engineering / research
Investigate the frontier of AI infrastructure
Design credible experiments and study compute allocation, numerical behavior, compilers, energy, and reproducibility.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 research
Investigate the frontier of AI infrastructure
Design credible experiments and study compute allocation, numerical behavior, compilers, energy, and reproducibility.
- 01
Experiment design and statistical power
Turn a systems performance claim into an experiment with a meaningful effect size, independent units, and an uncertainty budget.
32 min - 02
Benchmark design and contamination
Build comparisons whose workload, quality threshold, measurement boundary, and exposure history support the intended claim.
31 min - 03
Scaling fits and compute allocation
Use empirical scaling models to make constrained allocation decisions while testing extrapolation, fit uncertainty, and deployment costs.
34 min - 04
Communication-avoiding training
Evaluate accumulation, compression, and local updates by their effects on synchronization, numerical work, and time to a quality target.
35 min - 05
Numerical stability and mixed precision
Reason about range, rounding, accumulation, conditioning, and loss scaling before trusting a faster numerical training path.
35 min - 06
Compiler graphs and kernel fusion
Study graph capture, specialization, memory traffic, and fusion through measurable execution costs and correctness constraints.
34 min - 07
Distributed debugging and determinism
Localize the first divergence across workers while distinguishing randomness, nondeterministic execution, synchronization errors, and expected rounding.
35 min - 08
Energy and carbon accounting
Measure energy, define emissions boundaries, and compare AI systems using explicit functional units and uncertainty.
33 min - 09
Reproducible research artifacts
Package claims, exact inputs, execution environments, raw results, and analysis so an independent reader can verify the evidence.
32 min - 10
Systems research and open problems
Formulate testable infrastructure research questions, compare mechanisms fairly, and turn limitations into a disciplined next experiment.
36 min