Phase 19 — Interview Questions: Capstone — Maintainer & Startup
Staff/principal-level questions on this topic. Cover the answer, attempt it OUT LOUD, then compare. (See CAREER.md for how to run a full mock loop.)
Q1. How would you land your first vLLM contribution?
Model answer
Find a good-first-issue or a real bug you hit; run the duplicate checks (gh issue/pr search) per AGENTS.md; reproduce it; write a minimal fix WITH a test that pins the behavior and a clear PR description stating what you ran and why it isn't a duplicate; respond to review quickly. Specialize in one area to build reviewer trust over time.
Q2. Where's the moat for an inference startup built on vLLM?
Model answer
Not in renting GPUs around vanilla vLLM (margins compress). It's in a sustained kernel/scheduling edge, workload specialization (long-context/agentic/structured), the control plane (routing, autoscaling, multi-tenancy, cost attribution), or distribution/switching costs. Upstream commodity features; keep the genuine edge.
Going deeper
The flagship phases (02, 03) show the depth and number of questions to expect for a topic you claim as your specialty.