Learn ML engineering by building LLMs, tokenizers, transformers, vector databases, RAG systems, and more from scratch.
No credit card required.
Choose an ML system, work through its core pieces, and test your implementation as you build. All from your own development environment.
Build the machinery that turns raw text into tokens an LLM can process.
Pick an ML system you want to understand and build.
class BPETokenizer: def train(self, text, vocab_size): pairs = self.count_pairs(text) while len(self.vocab) < vocab_size: winner = max(pairs, key=pairs.get)Follow the steps and write the code using your own editor, terminal, and tools.
Push your code and see what works. Fix what fails and keep going.
Start with the fundamentals and work your way through the systems that power modern ML. New projects are added as the catalog grows.
Turn raw Unicode text into tokens, then train and optimize a byte-level BPE tokenizer.
Implement attention, normalization, training, and autoregressive text generation.
Build a computation graph, reverse-mode differentiation, and the machinery behind backpropagation.
Store embeddings, implement similarity search, and index vectors for fast retrieval.
Chunk documents, retrieve relevant context, and ground model answers in your own data.
Implement the noise schedule, denoising network, training objective, and image-sampling loop.
Pay $60 for complete beta access plus 12 months after beta ends, or pay $180 once for lifetime access. Both offers disappear when beta ends.
Build any project marked Free from beginning to end.
Use the complete catalog while Crackr is in beta.
Pay once and keep the complete catalog for good.
When beta ends
Public pricing
3 months
$60
1 year
$180
Lifetime
$499
Free projects stay free. Paid plans are one-time purchases with no automatic renewal.
Free projects include the complete curriculum and evaluations.
© 2026 Crackr. All rights reserved.
Built for engineers, by engineers.