Skip to content

Study to the depth you'll be tested at.

Six-stage sessions on hard ML and SWE topics: the intuition, the derivation, code that runs, and a retrieval check you have to answer. Add a resume and that check asks about work you have actually shipped, not a generic example. The tutor stops after every stage, and every thread saves so you can resume days later.

Two full sessions with no account, no card, no email.
157 topics · 10 domains · 2 tracks · 6 stages per session.
Attention weights, causal maskquery 4 / 12
softmax( QK^T / sqrt(d_k) ) Vstage 03 of Attention Mechanisms

Why this beats asking a chatbot to explain it

six differences
01

It stops and waits

A chat window hands you a wall of text and moves on. Here the session halts after every stage until you continue, ask, or take a card. That pacing is what turns reading into review.

3The mathcomplete
Waiting for you. Nothing arrives until you choose.
ContinueAsk a follow-upRevision card
02

The derivation is motivated

Every term gets a reason to exist. You learn what breaks when a term is removed, which is the question an interviewer actually asks.

3Scaled dot-product attention
Attention(Q,K,V)=softmax ⁣(QKdk)V\mathrm{Attention}(Q,K,V) = \mathrm{softmax}\!\left(\frac{QK^{\top}}{\sqrt{d_k}}\right)V

Why the scale factor. Drop it and the dot products grow with dimension:

Var(qk)dkLw0\mathrm{Var}(q \cdot k) \approx d_k \quad\Longrightarrow\quad \frac{\partial L}{\partial w} \to 0
03

The code runs

Reference PyTorch next to the math, with the mask in the right place and a test at the end. Theory and implementation on one screen.

4Implementationpython
def attn(q, k, v, mask=None):
    # scale before the softmax
    s = q @ k.transpose(-2, -1) / q.size(-1) ** 0.5
    if mask is not None:
        s = s.masked_fill(mask == 0, float('-inf'))
    return torch.softmax(s, dim=-1) @ v
mask applied before softmax, not after
04

You have to retrieve it

Stage 6 makes you reproduce the concept unaided, then grades the gaps. Recognition feels like knowing; retrieval is knowing.

6Retrieval check
State the attention equation and say what the scale factor does.
You scale by dk\sqrt{d_k} so the values stay small before the softmax.
+Correct. The equation and the scale factor are both right.
!Missing. Why saturation kills the gradient, not just that it keeps values small.
05

You write it, then it gets reviewed

The practice lab is an editor where you implement the thing from scratch in PyTorch or NumPy. What comes back is a code review — what is wrong and on which line, what you got right, and one question about a design choice you made.

Practice labyour code
def attention(q, k, v):
    s = q @ k.T / d_k ** 0.5
    return torch.softmax(s, -1) @ v
!Correctness. k.T transposes a 2-D tensor only — batched heads need transpose(-2, -1).
+What's right. You scaled before the softmax, not after.
?One deep question. You masked after computing every score. At a 32k context, what does that cost you?
06

It asks about your projects

Add a resume and the retrieval check stops quizzing you on a generic model. It asks you to defend the decisions you actually made and shipped — which is the version of the question an interviewer will ask.

6Retrieval checkfrom your resume
without a resume
State the attention equation and say what the scale factor does.
with your resume
Your two-tower retrieval service runs at a 30ms budget. Which attention layer would you drop first to hold it, and what degrades when you do?
Watch one run

A full session, exactly as it renders.

A recorded replay of all six stages — the same math, code, and diagrams a live session produces. No account, no model calls.

Demo session · Attention Mechanisms
Scripted replay · no model calls

Try it right here

Watch a full session run.

A recorded replay of a real six-stage session — the exact pauses where the tutor stops and waits, plus two follow-ups you can ask. No account, no model calls.

6 stages · the same math, code, and diagrams a live session renders · ~3 minutes

Six stages, in order, never compressed — the pause after each one is the point.

The format

Every topic, in the same six stages.

Never compressed, never reordered. The pause between stages is the product.

01
Big Picture
The concept in context: what problem it solves, where it appears in real systems, and the mental frame to keep before details arrive.
context
02
Intuition + Visual
Core idea in plain language, then a structured diagram with tensor dimensions and data flow annotated. Mandatory for all DL architectures.
diagram
03
The Math
Step-by-step derivation with every term motivated. Not just what each symbol is, but what breaks if you remove it.
derivation
04
Implementation
Production-quality PyTorch with type annotations, every non-obvious line commented, and an explicit test snippet at the end.
PyTorch
05
Interview Questions
5 calibrated questions: conceptual, implementation, applied, systems-level, and failure modes. Interview prep is the first proving ground.
5 questions
06
Retrieval Check
Retrieval drill. The AI asks, waits for your answer, then tells you precisely what was right, wrong, or missing.
you answer

ML / Research

71 topics · 5 domains

Attention, transformers, optimization, RLHF, training systems, and the math behind modern models.

Classical ML / Deep Learning / Reinforcement Learning / Training Engineering / Systems and MLOps
for Research Engineer · MLE · Research Scientist · Applied Scientist

Software Engineering

86 topics · 5 domains

Rendering models, distributed systems, data structures, caching, databases, and production tradeoffs.

Frontend / Backend / System Design / UI/UX / CS Fundamentals
for Frontend · Backend · Fullstack · System Design · UI/UX
Read one first, no sign-up

17 concepts written in the session format.

Browse all 17
Access

Open by default. Sign in for the parts that have to know you.

Read every concept, browse all 157 topics, and run 2 full six-stage sessions before you decide anything. An account raises that to 8 a month and turns on the parts that cannot work without knowing who you are.

Session historyReload any thread in full and continue from the exact stage you stopped at, days later.
Revision cardsA compact card per topic: core concept, key equations or tradeoffs, and the highest-value follow-ups.
Profile contextBackground and focus areas shape each session, so explanations land at your level instead of a generic one.
Grounded retrievalAdd a resume and stage 6 stops asking about a generic model — it asks you to defend the implementation choices you actually made.
Practice labImplement the concept from scratch and get a line-by-line review: what is wrong, what is right, and one question about a choice you made.
Profileneeds an account
Resume
Stage 6 asks you to reason about work you shipped, not a textbook example.
optional · deletable
STAR stories
Generated from your own work, not invented examples.
from resume
Focus areas
Shapes which tradeoffs each session leans on.
editable anytime
Target roles
Calibrates stage 5 questions to the interviews you want.
MLE · Research Eng

Resume upload is optional, skippable, and deletable at any time. It exists so STAR stories and follow-ups can reference work you have actually shipped.

Sign in to set up context
Catalog

71 topics, 5 ML domains.

Classical ML
16
Linear Regression
Logistic Regression
Support Vector Machines
K-Nearest Neighbours
K-Means Clustering
Naive Bayes
Decision Trees
Random Forests
+ 8 more
Deep Learning
26
Backpropagation and Computational Graphs
Activation Functions
Loss Functions
Optimizers (Adam, AdamW, LR schedules)
Batch Normalization and Layer Normalization
Weight Initialization
Feedforward Networks
CNNs
+ 18 more
Reinforcement Learning
14
Markov Decision Processes
Bellman Equations
Value Functions
Policy-based vs Value-based vs Model-based RL
Dynamic Programming
Monte Carlo Control
Temporal Difference Learning
Q-Learning
+ 6 more
Training Engineering
8
Gradient Accumulation
Gradient Checkpointing
Mixed Precision Training (fp32, fp16, bfloat16)
CUDA Setup and Device Management
Distributed Training and DDP
Training Config and YAML Structure
Multi-Seed Evaluation and Reproducibility
Experiment Tracking with MLflow
Systems and MLOps
7
Model Serving and Deployment
Embeddings at Scale
Vector Databases
Batch vs Online Inference
Latency vs Throughput Tradeoffs
CI/CD for ML
Experiment Tracking
Pricing

Try it before you exist to us.

Only Pro takes a card. The trial does not charge for seven days and cancels in one click.

Guestno account
$0
2 full sessions, then sign in
Two complete six-stage sessions
All 157 topics, all 17 concept pages
No email, no card, no profile
No practice lab, nothing saved between visits
Freesign in
$0
8 sessions a month
8 sessions a month, no card
Saved threads you can resume
5 practice-lab code reviews a month
Revision cards
Stage 6 grounded in your resume, not generic examples
Pro7-day trial
$9 / month
Unlimited sessions
Unlimited review sessions
Unlimited practice-lab code reviews
Everything in Free
New topics ship to Pro first
Request topics directly

Pick a topic. Leave able to reproduce it.

Start a sessionTwo sessions with no account, 8 a month once you sign in