AI technical screening
Your first-round technical interview, run to a published standard.
Live by Technical Interviews conducts a real conversational interview, by voice, with a shared code editor or a system design whiteboard. Every candidate is scored against the same four-dimension rubric from the Interviewer’s Guide, at the level you are hiring for, with evidence you can read.
3 interviews free · no credit card · about 20 minutes to your first invitation
The bottleneck is not interviewer availability. It is interviewer consistency.
Every engineering org already knows screening is expensive. Fewer notice that it is also inconsistent enough to make some of their hiring decisions close to random.
Four steps, then you are reading evidence.
Create the job
Title, team, level (Junior/Mid, Senior, or Lead/Staff+), and which interviews to run. Every interview is conducted by Bracket, our one interviewer. The level is not cosmetic: it changes the bar the interviewer holds and the weights the score is computed with.
Add your candidates
Type them in or upload a CSV. Pulling them from Greenhouse, Lever, or Ashby is planned, not built yet. Each one gets an emailed link. No account, no password, no download.
They interview
A real spoken conversation, 25 to 45 minutes, with a shared code editor or a system design whiteboard. The interviewer asks, waits, probes on depth, changes a constraint, and works to the time structure the book recommends.
You read the evidence
Four dimension scores, each with a written rationale and the quotes it rests on, every quote checked against the transcript. A timeline that jumps to each cited moment, with the transcript beside it. You decide. Live never rejects anyone.
The rubric is published. You can read it before you buy anything.
Live scores against the four-dimension competency model from Crushing the Technical Interview: Interviewer’s Guide, written by Keith Henning, a director of enterprise architecture with 30 years in software and system design.
It is not a theory that was written down once. These four dimensions and their anchors have been implemented in real hiring loops at several companies over more than a decade, and revised each time they produced an answer the hiring managers could not defend. The book is the current state of an instrument that has been tuned against outcomes, not a proposal.
Behavioral anchors at every score. Explicit calibration for junior, senior, and staff. No proprietary black box, no unexplained five-point scale. When your legal team asks what the system assesses, you hand them a book.
Read the full rubric →Problem Decomposition
Turning a vague prompt into a scoped, workable problem. Asking before building. Separating requirements from assumptions.
Architectural Judgment
Reasoning through trade-offs rather than pattern-matching. Not which choice they make, but whether they can name what they are accepting.
Technical Depth
Going deep on demand. Broad sketches are cheap. Rehearsed knowledge collapses fast when you push on a single component.
Communication
Thinking out loud while deciding, not after. Naming assumptions under ambiguity. Scored on reasoning conveyed, never on style, accent, or confidence.
A real whiteboard, not a text box.
Async assessment platforms cannot run a system design interview, because system design is a conversation over a drawing. Live gives the candidate an infinite canvas with an architecture stencil set, and an interviewer who asks about the design as it takes shape.
Typed edges
Connections carry protocol and direction: sync request, async publish, replication, read path, write path. That is what turns a picture into a design.
Requirements, tracked separately
Functional and non-functional requirements, assumptions, and anything ruled out of scope live in their own pane, not buried in the diagram.
Semantic export
Every canvas serializes to a typed node and edge graph, not pixels. A design is read as structure, never as a picture.
Every score cites a moment. Click it and read it in context.
A quote that does not match the transcript word for word is discarded, and when too few verified quotes are left the report says insufficient signal instead of giving a score. The system is not permitted to have an opinion it cannot support.
No name or contact details are sent to the evaluator alongside the transcript or, for System Design, the final whiteboard. That is the strongest single bias mitigation available, and it is not optional.
Book thirty minutes and we will run one live →Architectural Judgment
4.0Rationale: names the trade-off being accepted for most significant choices, tied to stated requirements.
We route your attention. You draw the conclusion.
An estimate of whether a candidate had outside help is planned and not built yet. When it ships it will show the specific, timestamped observations behind the estimate, it will not accuse, it will never change a score, and it will never reject anyone. Today, Live does no facial analysis, emotion inference, or gaze tracking, because those are unreliable and unfair.
How assistance signals are planned to work →Built for the questions your legal team will ask.
Jurisdiction-aware notice and consent for Illinois, Maryland, Colorado, and New York City. Candidates can decline automated evaluation and request a human interview without penalty. SOC 2 Type II is targeted, not completed yet. GDPR and CCPA aligned. Candidate data is never used to train any model, by us or by any vendor. Read the security and privacy detail.
Keith Henning
Director of Enterprise Architecture with roughly 30 years in software and system design, and author of the Crushing the Technical Interview series: Data Structures and Algorithms across four languages, System Design, and the Interviewer’s Guide that Live’s rubric comes from.
Live exists because the book makes an argument (that interview failure modes are structural, and structural problems have structural remedies) and software is how you actually apply a remedy at scale.
Priced against what a screen actually costs you.
A screen consumes about 75 minutes of senior engineer time. Live starts at $25.
Unlimited jobs and users. Priced for hiring below the volume Starter is built for.
One team, unlimited jobs and users. Expensable without a procurement conversation.
Unlimited jobs and users. Planned for this plan, not built yet: ATS integration, org rollups, API and webhooks.
Planned for this plan, not built yet: multiple hierarchies, SSO, voice choice, rubric weight adjustment, bias audit support.
Send one to yourself first.
Run an interview, then read your own scorecard with the evidence links. It is the fastest way to know whether this is good enough to send to a candidate you care about.