Skip to content
mlmentorship

Front matter

Read by book or follow one focused route.

The table of contents is the main path. Use one route below only when you need to cross several books for a specific goal.

Start here

Three lessons, then choose your depth.

Use this 30-minute route to see how the field guide teaches. Close each lesson and explain the mechanism before moving on.

  1. 01

    See a mechanism

    Build spatial intuition for a matrix before touching the formula.

    Read the visual, close it, and explain what changes shape.
  2. 02

    See a systems trade-off

    Follow memory movement and tiling instead of memorizing a kernel name.

    Explain the bottleneck and one condition that would change the design.
  3. 03

    Practice interview judgment

    Work through a real trade-off, then defend the decision under pushback.

    Answer closed-book before reading the leveled response.

Ready for a real loop? Build the personalized map →

01

Build deep technical foundations

Use this for a compact pass through math, learning, and transformer basics.

  1. Linear algebra: linear maps, SVD and PCA, eigenvalues.
  2. Probability: maximum likelihood, Bayes rule, KL divergence.
  3. Deep learning: backpropagation, attention, transformers.
  4. Evaluation: hypothesis testing, fair comparison, evaluation validity.
  5. Use the concept index only when an attempt exposes another gap.
02

Study large-model systems

Use this for training, inference, accelerator, or Research Engineer loops.

  1. Count parameters, FLOPs, and memory.
  2. Derive communication from sharded tensor axes.
  3. Map the layout to the accelerator network.
  4. Choose a parallel plan and measure MFU.
  5. Attempt the worked 70B training plan, then read distributed profiling.
03

Move from research to industry

Use this when strong research experience must become clear interview evidence.

  1. Compare applied, engineering, and research role expectations.
  2. Practice fair paper critique.
  3. Design a discriminating ablation.
  4. Prepare the ambitious-project discussion.
  5. Present one project under interruption and pushback.
04

Calibrate a senior ML hire

Use this to define the role and evaluate evidence at the intended level.

  1. Define the role before writing the loop.
  2. Choose the capabilities the loop should test.
  3. Calibrate senior through senior-principal evidence.
  4. Choose representative prompts from the question index.

← Back to the full library