Coding problems · in-browser

Small, sharp, self-contained

Every problem runs in your browser via Pyodide. No submission queue, no leaderboard, no login. The tests are on the page — read them, then write until they pass.

• Cosine similarity from scratcheasy

  The distance function behind every vector search and RAG retriever. Two vectors in, one number in [-1, 1] out — and the edge cases are where interviews live.

  Retrieval

• Implement softmax (numerically stable)easy

  Turn a vector of scores into a probability distribution — the function at the end of every classifier and inside every attention head. The catch is doing it without overflowing.

  Neural network components

• Scaled dot-product attentionhard

  The single operation at the heart of every transformer. Implement it from scratch in pure Python — matmul, scale, row-wise softmax, matmul — and the mechanism stops being a mystery.

  LLM internals