- Length
- 55 min
- Format
- One engineer on video with screen share, usually in CodeSignal (Google Colab for research roles), Python by default; one practical problem that grows in parts
- They grade
- Working code, how you reason out loud, how you absorb new requirements
- AI use
- No AI; looking up documentation and searching the web is fine
- Heads-up
- Problems are practical components (for example stack traces, file deduplication, a concurrent crawler, a tokenizer), not puzzles. The base problem takes about a third of the hour; follow-ups on scale, IO versus CPU and parallelism fill the rest. The recruiter often describes the problem's theme in advance. Research roles choose from a menu instead (ML programming, coding and design, RL debugging, prompting in a notebook, or a take-home); some screens are an agent loop with tool use.
