Length 30 minFormat CallThey grade Background, level, location and logistics
Length 45 minFormat Call with the hiring managerThey grade A specific reason you care about legal AI, role fit and past workHeads-up General excitement about frontier models falls flat.
Length 60 minFormat Live coding in a shared editorThey grade One medium problem (arrays, strings, hash maps, sometimes parsing), working and tested code, reasoning out loudHeads-up The set of problems is small and stable: tagging citations or highlights in text, an in-memory file system, a spreadsheet with formulas. Recruiters often hold a prep call first.
Length 60 minFormat Live practical codingThey grade Structure, edge cases and how you take hints; a follow-up changes a constraintAI use Candidates report freedom to use the tools and languages they know; confirm with your recruiterHeads-up Reported problems include finding every occurrence of a phrase in a long document and merging intervals. Interviewers run their own, sometimes hidden, tests, and recent loops expect bug-free code. ML roles get a Colab round on a small RAG pipeline instead.
Length 60 minFormat Design discussion on a whiteboard or docThey grade Retrieval, evaluation, multi-tenancy, confidentiality between clients and faithful answers with citationsHeads-up A reported prompt: answers over a law firm's private documents that cite the exact clause. Other loops get a document store with organization-level access control (like Harvey's Vault) instead, so prepare a storage design as well as a RAG design. Often held in the San Francisco office with the project walkthrough, while coding and values run over video.
Length 45 minFormat ConversationThey grade Working with lawyers and other non-engineers, ambiguity, ownership, shipping fast
Length 60 minFormat Walk through a project you built, then questionsThey grade The tradeoffs you personally owned, probed until the edge of what you knowHeads-up Candidates underrate this round; vague answers fail it fastest.
