Learn ML by building it from scratch.
Implement algorithms, gradients, and neural architectures using pure vectors, matrices, and math. No frameworks. No shortcuts.
No credit card required
Try Solving a Real ML Statistics Problem
No datasets. No GPUs. Just vectors, math, and code.
LINEAR REGRESSION I
Implement linear regression prediction using vectorized matrix operations. Compute ŷ = XW + b for an input batch.
GRADIENT DESCENT STEP
Compute the parameter update for a single gradient descent step given weights, gradients and learning rate.
RELU ACTIVATION FORWARD
Implement the forward pass of a ReLU activation layer for a batch input vector.
SCALED DOT ATTENTION
Implement scaled dot-product attention used in transformer architectures.
SOTA_RESEARCH
IMPLEMENT PAPERS THE
MATH WAY
Reading a paper is the first 10%. The real understanding happens when you translate equations into compute.
Axiom-ML breaks down seminal works like "Attention Is All You Need" into granular coding milestones. You don't just learn about self-attention; you implement tensor operations, scaling factors, and masking logic from scratch.
RESEARCH GOAL
CONSTRUCT A FULL GPT-STYLE DECODER ARCHITECTURE.
MILESTONE_01 // COMPLETED
TOKEN & POSITIONAL ENCODING
Generate sinusoidal embeddings to provide sequence order information to the model.
MILESTONE_02 // COMPLETED
MULTI-HEAD ATTENTION (MHA)
Parallelize attention mechanisms across multiple projection subspaces.
MILESTONE_03 // IN_PROGRESS
LAYER NORM & FEED-FORWARD
Stabilize training with layer normalization and point-wise fully connected networks.
MILESTONE_04 // LOCKED
THE TRANSFORMER ENCODER/DECODER STACK
Assemble the full architecture and implement the residual connections.
01 // FUNDAMENTALS
FOUNDATIONS
Build the mathematical foundations behind machine learning. Instead of relying on libraries, Axiom walks through the derivations of core algorithms from linear algebra and calculus and then guides you to implement them yourself.
02 // LEVEL UP
RESEARCH
After mastering the fundamentals, dive into research implementations. Recreate influential ML papers from scratch and understand how modern architectures actually work under the hood.
03 // LEVEL UP (QUITE LITERALLY)
PROGRESSION
Learning is gamified through streaks, milestones, and model challenges. Train models, complete modules, and unlock higher level systems while maintaining consistency in your learning journey.
04 // HOST
INTERVIEW
Axiom also hosts machine learning interviews directly on the platform. Recruiters and candidates can meet in a shared environment where ML systems, derivations, and implementations can be explored live.
Master Machine Learning From First Principles
Start solving our vector-based problems for free.
Upgrade to unlock courses built around landmark research papers.
BASIC
$0 / month (Free)
- ✓ Access to Core Problem Library
- ✓ Participate in Public Contests
- ✓ Global Leaderboards
- ✕ Some Paper Implementation Tracks
ELITE
$9 / month
EXCLUSIVE_POWER_UPS
- All Paper Implementation Tracks
- All Courses
- Host Interviews as Interviewer
- Early Access to Latest Courses
SO WHY WAIT?
BUILT FOR THOSE WHO
REFUSE TO TREAT MODELS
AS A BLACK BOX
This is not another course platform. It's a system for engineers who want to break models apart, inspect every tensor, and rebuild them from scratch.
20+ ACTIVE
STATUS
[✓] CORE PROBLEMS UNLOCKED
[~] PAPER TRACKS AVAILABLE
[?] YOU HAVEN'T STARTED
NO CREDIT CARD • START FREE