Find the top three similar documents instantly
one move: search by similarity
Open the code pane immediately, then use the Spec tab, optional design question, tests, and artifact when you want them.
Finish each move before carrying the artifact forward.
A vector index trades exhaustive scans for a data structure built for nearest neighbors.
Deliverable: Write the mechanism in one sentence.Embed a small document set and build an exact inner-product index.
Deliverable: Type the smallest runnable implementation.Top-k results must be stable, normalized, and tied to source document IDs.
Deliverable: Record the expected output and one edge case.Change k, add distractors, and compare exact search with a brute-force check.
Deliverable: Change one variable and explain the result.A versioned vector index plus a retrieval-quality sample.
Deliverable: Save the code, output, and a short failure note.Top-k results must be stable, normalized, and tied to source document IDs.
Change k, add distractors, and compare exact search with a brute-force check.
A versioned vector index plus a retrieval-quality sample.
Mapped from the supplied Build Everything PDF, source page 40. The five moves are Deriva’s implementation contract for this project.