Azimuth

The Paths

Pick a destination — the sky draws the route in gold.

AI Engineer

A working map of the whole field — enough depth in each pillar to build with any of them.

For engineers who ship — the ideas you will actually deploy, in the order they compose.

15 papers≈ 8 weeksstarts foundational, ends intermediate

  1. Backpropagation
  2. CNN (LeNet)
  3. ResNet
  4. word2vec
  5. Transformer
  6. Scaling Laws
  7. GPT-3
  8. InstructGPT
  9. REINFORCE
  10. DQN
  11. CLIP
  12. Latent Diffusion
  13. Whisper
  14. RAG
  15. ViT

Computer Vision Engineer

From backprop to Vision Transformers — the papers behind every vision system you will ship.

For builders of systems that see — from filters to features to full perception stacks.

14 papers≈ 7 weeksstarts foundational, ends intermediate

  1. Backpropagation
  2. CNN (LeNet)
  3. ImageNet
  4. AlexNet
  5. VGG
  6. BatchNorm
  7. ResNet
  8. Faster R-CNN
  9. YOLO
  10. U-Net
  11. Transformer
  12. ViT
  13. CLIP
  14. Segment Anything

LLM Engineer

The lineage from attention to GPT — understand the machinery you will fine-tune and deploy.

For builders who want working systems — you leave able to reason about every layer of a modern LLM.

15 papers≈ 8 weeksstarts foundational, ends intermediate

  1. Backpropagation
  2. word2vec
  3. Seq2Seq
  4. Attention
  5. Transformer
  6. BPE
  7. GPT-1
  8. BERT
  9. Scaling Laws
  10. GPT-3
  11. InstructGPT
  12. LLaMA
  13. LoRA
  14. FlashAttention
  15. RAG

ML Researcher

Read the field the way it was written — full lineage, original problems, open questions.

For readers who want the why — the intellectual spine of the field, argument by argument.

15 papers≈ 10 weeksstarts foundational, ends intermediate

  1. Backpropagation
  2. Bias–Variance
  3. CNN (LeNet)
  4. Dropout
  5. BatchNorm
  6. REINFORCE
  7. Attention
  8. ResNet
  9. Transformer
  10. GPT-1
  11. BERT
  12. Scaling Laws
  13. Chinchilla
  14. Double Descent
  15. Grokking