Algorithm Researchers

Tick units as you finish them. Progress is saved only in this browser — no account needed.

1. Python, NumPy & Tooling

   Stage 1 · 12 h · 12 h total

   Read and write Python, vectorise with NumPy, use the command line, Git and Jupyter.

   I finished this

2. Linear Algebra

   Stage 1 · 15 h · 27 h total

   Vector spaces, matrix multiplication, eigenvalues and SVD — the language of embeddings, attention and LoRA.

   I finished this

3. Calculus & Optimization

   Stage 1 · 12 h · 39 h total

   Derivatives, partial derivatives, the chain rule and gradients — “walking downhill along the gradient”.

   I finished this

4. Probability, Statistics & Information Theory

   Stage 1 · 12 h · 51 h total

   Distributions, expectation, Bayes, maximum likelihood, cross-entropy and KL divergence.

   I finished this

5. PyTorch & Autograd

   Stage 1 · 10 h · 61 h total

   Tensors, autograd, data loading and the training loop.

   I finished this

6. The ML Workflow: Data, Loss & Generalization

   Stage 2 · 10 h · 71 h total · lesson ready

   Train/validation/test splits, overfitting, bias–variance, cross-validation and metrics.

   I finished this

7. Unsupervised Learning

   Stage 2 · 8 h · 79 h total

   k-means, Gaussian mixtures and EM, PCA, t-SNE/UMAP.

   I finished this

8. Neural Networks & Backpropagation

   Stage 3 · 10 h · 89 h total · lesson ready

   From neurons and multilayer perceptrons to backpropagation and activation functions.

   I finished this

9. Training Deep Nets

   Stage 3 · 10 h · 99 h total

   SGD/momentum/Adam/AdamW/Muon, learning-rate schedules, BatchNorm/LayerNorm/RMSNorm, dropout and residuals.

   I finished this

10. CNNs, RNNs and the Birth of Attention

    Stage 3 · 10 h · 109 h total

    Convolutional networks, LSTMs, Seq2Seq, word vectors and Bahdanau attention.

    I finished this

11. Reinforcement Learning Basics

    Stage 3 · 14 h · 123 h total

    MDPs, Q-learning, DQN, policy gradients, PPO and AlphaZero — the prerequisites for post-training and reasoning models.

    I finished this

12. Tokenization & Embeddings

    Stage 4 · 6 h · 129 h total · lesson ready

    BPE tokenisation, per-token pricing, embeddings and semantic similarity.

    I finished this

13. Transformers & Attention

    Stage 4 · 16 h · 145 h total · lesson ready

    Self-attention, multi-head attention, positional encoding, residuals and normalisation — build a GPT from scratch.

    I finished this

14. Modern LLM Architecture

    Stage 4 · 12 h · 157 h total

    RoPE, SwiGLU, RMSNorm, GQA/MLA, FlashAttention, sliding windows and context extension.

    I finished this

15. MoE & Beyond-Transformer Architectures

    Stage 4 · 10 h · 167 h total

    Sparse experts, load balancing, Mamba/SSMs, linear attention and hybrid architectures, diffusion language models.

    I finished this

16. Pretraining, Data & Scaling Laws

    Stage 4 · 10 h · 177 h total

    Pre-training data pipelines, scaling laws, Chinchilla, μP, MTP and training stability.

    I finished this

17. Instruction Tuning & PEFT

    Stage 5 · 10 h · 187 h total · lesson ready

    SFT, synthetic data, LoRA/QLoRA/DoRA, adapters and model merging.

    I finished this

18. Alignment: RLHF, DPO & Constitutional AI

    Stage 5 · 10 h · 197 h total

    Reward models, PPO-style RLHF, the DPO family, RLAIF and reward hacking.

    I finished this

19. Reasoning Models: RLVR, GRPO & Test-Time Compute

    Stage 5 · 12 h · 209 h total

    Chain of thought, self-consistency, process rewards, GRPO/DAPO, verifiable rewards and the R1 recipe.

    I finished this

20. Quantization, Pruning & Distillation

    Stage 6 · 8 h · 217 h total

    INT8/INT4, GPTQ, AWQ, SmoothQuant, BitNet and knowledge distillation.

    I finished this

21. GPUs, Distributed Training & AI Chips

    Stage 6 · 10 h · 227 h total

    Memory, bandwidth and compute; mixed precision; data/tensor/pipeline parallelism; ZeRO/FSDP; the chip landscape.

    I finished this

22. Diffusion, Flow Matching & Image Generation

    Stage 7 · 14 h · 241 h total

    VAE/GAN, DDPM/DDIM, score SDEs, latent diffusion, CFG, DiT, flow matching and ControlNet.

    I finished this

23. Interpretability: from Attribution to Circuits

    Stage 9 · 10 h · 251 h total · lesson ready

    Attribution, probes, the logit lens, superposition, SAEs, circuit tracing, knowledge editing and activation steering.

    I finished this

24. Tracking the Frontier: Reading & Reproducing Papers

    Stage 9 · 6 h · 257 h total

    The three-pass reading method, tracking arXiv and conferences, reproduction and writing.

    I finished this

• 1Everyday Users · No code: in 2–3 weeks, learn to understand and use AI well and spot hallucinations and risks.40h
• 2Application Developers · Python is enough: in 3 months, ship a production-ready RAG / agent app you can evaluate and deploy.126h
• 3Product Managers & Leaders · Understand technical limits, cost structure and risk well enough to choose tools, start projects and set governance.84h
• 4Domain Scientists · With a background in your own field, learn to do research with ML and LLMs — linked with the “AI × Disciplines” section.143h