A cloud loop is four to six stages: a phone screen of rapid-fire fundamentals, a hands-on or take-home round that is increasingly the core of the process, a 60 minute design round judged on resilience and cost rather than algorithms, and a behavioural incident round. GhostPilot runs in the browser side panel for the live ones.
Free tier: 10 minutes of live interview time a week, no card. Windows desktop app for full-screen share.
A trap question, and the wrong instinct is to answer about the apply rather than the plan. The structured answer surfaces about two seconds after the question ends: read the diff for the forcing attribute, reach for create_before_destroy, consider state mv or an import rather than an apply, and note that a Multi-AZ recreate is a different conversation about RTO and RPO. Live transcription then holds the follow-up ("now make it fifteen engineers across three environments, where does state live") so the S3 plus DynamoDB locking half does not get dropped.
"An EC2 instance in a private subnet cannot reach the internet." "Production latency just tripled, walk me through your first ten minutes." "The monthly bill jumped forty percent with no traffic change." Three questions that sound near-identical for their first four words and want three different ordered methods: route table, NAT gateway, NACL, security group; the four golden signals then recent deploys; Cost Explorer with cost-allocation tags, then data transfer. Question detection catches which one was actually asked.
"An application needs read access to one S3 bucket. How do you grant it?" The answer they are listening for involves a role and a scoped policy; the answer that ends the round is s3:* on Resource: *. Behavioural questions are handled differently, so "tell me about the worst production incident you were part of" comes back shaped like a blameless postmortem, with blast radius and what you changed afterwards, rather than a list of services that were on fire.
GhostPilot listens to call audio, so an offline take-home where you write a Terraform module on your own machine is outside what it does. It is for the live rounds: the phone screen, the design round, the incident round, and the walkthrough afterwards.
The accurate version, because a wrong answer here costs you the interview. The Chrome extension lives in the browser side panel: it is not captured when you share a single tab, nor when you share a window with the panel popped out. Like every extension, it is visible if you share your entire screen. The Windows desktop app is different. It excludes its own window at the operating-system level, so it stays out of screen shares, recordings and screenshots, and sits in the taskbar as "GP Helper".
Hands-on cloud rounds often run in a shared browser console tab, and a tab share does not capture the side panel. If your interviews force a full-screen share, the desktop app is the right tool, and the only one we would stand behind.
Point it at a mock design round
Ten minutes of live interview time a week, free, no card. Enough to check how it handles a scenario question with a trap in it.
Install on Chrome