What to screen for
The first technical round usually tries to answer one question: is this person worth an hour of an engineer's time? Answering it with a coding puzzle tests a narrow skill and, today, mostly tests whether someone pasted the problem into a chatbot. Asking people to talk through real decisions tells you more.
- Technical depth. Do they understand why a tool or pattern works, and where it fails, or are they repeating names?
- Tradeoff judgment. Asked "why X over Y", do they name both sides without being prompted?
- Correctness. Would the design or fix they describe actually work?
- Communication. Can they explain a system to a colleague clearly, at the right level of detail?
These are the competencies on the default scorecard of Atlas, Nova's engineering profile: Technical depth (30%), Tradeoff judgment (25%), Correctness (25%) and Communication (20%). You can change them per job.
Example questions
Examples
These were written for this page. For your job, Infyva generates questions from the job description and the candidate's resume if you provide it. On Pro and Enterprise you can use a fixed question bank instead.
An API endpoint that used to respond in 80 milliseconds now takes two seconds, but only for some customers. How do you find out why?
Narrows down what the slow customers have in common, checks metrics and traces before guessing, and names likely causes (data size, a missing index, a noisy neighbor) with how to confirm each.
You need to process uploaded files that can take several minutes each. Would you do it inside the request or in a background job? Why?
Chooses a background job and explains the costs: retries, idempotency, how the user learns the job finished, and what happens when a worker dies mid-file.
Tell me about a technical decision you made that turned out to be wrong. What did you learn?
A real decision, the information they had at the time, how they found out, and what they changed. Blaming others is a weak answer.
How would you design rate limiting for a public API used by thousands of customers?
Picks an algorithm (token bucket, sliding window), says where state lives, handles multiple servers, and thinks about what limited clients see.
Explain to a product manager why a feature they want will take three weeks instead of three days.
Plain language, the specific work involved, and an option for a smaller first version.
How scoring works
Every answer is transcribed and scored against the scorecard, with written reasoning next to the recording. A competency score cannot sit more than 15 points above the answer it cites as evidence, so the scorecard cannot run ahead of what the candidate actually said. If fewer than three questions are answered, the report says there is not enough evidence to recommend a next step.
Speaking style, accent and tone of voice are not scored. That is a real advantage over live screens for engineers who are strong but not polished speakers.
Integrity signals
During the interview Infyva records browser signals: leaving the interview tab, pasting, and, if the candidate opts in, a face presence check that counts how many people are in view, in the browser. Each signal is tied to the answer it happened on. Open that answer, listen, and decide. A tab switch during a question about a past project means something different from one in the middle of a design question, and a person is better placed to judge that than a rule.
Adding a coding round
On Pro and Enterprise you can add a vibe-coding round for technical roles. The candidate builds something with an AI coding agent in a secure sandbox, and you see how well they work with AI: how they direct it, check its output and fix what it gets wrong. For many teams that is closer to the actual job in 2026 than a whiteboard puzzle.
What the report shows
- A recording and transcript of every answer, so a senior engineer can listen to the one design question in two minutes.
- Per-competency scores with reasoning, strengths and concerns.
- Integrity notes on the answer they happened on.
- A share link that hiring managers open without an account, and Greenhouse sync on Pro and Enterprise.
For more on why consistent questions matter, read our structured interviews guide, and for comparing tools, the AI interview software overview.
Frequently asked questions
Does the voice interview include live coding?
The voice interview covers spoken technical depth, tradeoffs and debugging approach. Turn on coding questions when you generate the interview and those questions open a code editor. On Pro and Enterprise you can also add a vibe-coding round, where the candidate builds with an AI coding agent in a secure sandbox.
How do you handle candidates using AI to answer?
Infyva records browser signals during the interview, such as leaving the tab and pasting, and ties each one to the answer it happened on. You can play that answer back and decide. Flags are evidence for a person to judge, not an automatic fail.
Which competencies does the engineering profile score?
The engineering Nova profile ships with Technical depth (30%), Tradeoff judgment (25%), Correctness (25%) and Communication (20%). You can change them per job.
Can I connect it to Greenhouse?
Yes, on Pro and Enterprise. Candidates import from Greenhouse and results go back to the candidate record.