Screen for real ability, with evidence.
Nura, our AI interviewer, runs a natural, role-specific first-round interview, then AskNaru hands your team a report where every score links to the exact artifact, action, and transcript moment. Decision support your hiring managers can actually audit.
Structured interview engine
Manages the agenda, timing, difficulty, recovery, and follow-up policy, so every session is structured and comparable, not improvised.
Real-time conversational voice
Natural speech with barge-in and turn-taking. Nura reacts to what the candidate says and does, and probes exactly where it matters.
Versioned rubrics
Role, level, and company-specific rubrics, versioned and job-related. Change a rubric and every future score is tied to the exact version used.
A structured evidence graph, not a transcript and a number.
AskNaru's database retains a full evidence model. Each item ties a claim to its source, artifact, and timestamp, then to a competency, a score, and the evidence that supports or contradicts it.
- Calibrated to your barCompany, role, level, and rubric version, mapped to the employer's actual expectations.
- Traceable to the momentSemantic events, not raw keystrokes. A reviewer can jump straight to what happened.
- Confidence and human overrideReviewer agreement, override, and reason are captured alongside every score.
{ "role": "frontend_engineer", "target_level": "sde_2", "rubric_version": "fe-sde2-v4", "timeline": [ { "timestamp": "00:31:08", "source": "editor_and_tests", "event": "fixed stale-closure bug", "artifact_ref": "useSearch.ts:14", "competency": "problem_solving", "score": 3, "confidence": 0.88, "reviewer_override": null } ] }
Each adapter reads the evidence a role actually produces.
A generic model watching a shared screen misses what matters. Each adapter instruments the real work of the role.
Instrumented coding workspace
No screen recording required. A browser IDE with a secure, isolated runtime captures structured telemetry far more reliable than pixels: code snapshots and semantic diffs, hidden-test results, compiler and runtime errors, terminal and debugger actions, time between attempts, hints, and approved assistance, with spoken reasoning linked to the code that changed.
Portfolio & work-sample interviewing
Original portfolio files are uploaded before or during the session. Live frames are sampled with event-triggered high-resolution capture and OCR for dense case studies. Claims are linked to the exact artifact and transcript moment, and inconsistencies between what is said and what is shown become follow-up questions.
Instrumented diagram canvas
Candidates design on a canvas that records the reasoning, not just the final picture: components and relationships created and removed, assumptions, responses to new scale constraints, bottleneck analysis, and trade-offs, all preserved as evidence.
Structured, job-related probing
The behavioural layer uses structured voice probing with transcript evidence and consistency checks against a job-related rubric. It never infers emotion, personality, honesty, or competence from gaze, facial movement, or accent. Those pseudo-signals are excluded by design.
What your hiring team actually receives.
A report designed to be read by a human and defended in a debrief, not an opaque verdict. It recommends progression. It never says “definitely hire.”
- Progression recommendationStrong Yes, Yes, Borderline, or No, with the reasoning behind it.
- Competency scores mapped to JD & levelDeterministic test and execution results where applicable.
- Claims linked to evidenceEvery conclusion cites a transcript, action, or artifact, with confidence.
- Questions to validate nextWhat the human interviewer should confirm in the next round.
“The candidate passed 9 / 10 tests, diagnosed a concurrency issue without assistance, and explained the trade-off correctly. However, they did not consider idempotency until prompted. Recommended to progress, with distributed-systems experience to validate in the next round.”
Rules that keep the scores honest.
Facts before judgment
Deterministic tools, tests, execution, complexity checks, and static analysis establish what is true before the AI judges reasoning.
Multiple valid approaches
There is rarely one right answer. Different sound solutions can each receive full credit against the rubric.
Every score cites evidence
No conclusion stands without supporting evidence and a stated confidence level.
Job-related, versioned rubrics
Rubrics are role-specific, level-specific, and versioned, never a generic, opaque readiness number.
No pass-probability theatre
No claims about the odds of a hire succeeding until enough consented, company-specific outcome data exists.
No pseudo-signals
Emotion, personality, honesty, and competence are never inferred from gaze, facial movement, or accent.
An AI first round that makes the human rounds better.
AskNaru does not replace your process. It strengthens the front of it and hands richer evidence to the people who decide.
Resume & JD-specific interview
- Fundamental technical & behavioural questions
- Dynamic follow-ups on real answers
- Communication & knowledge-depth signal
Work-sample in the right workspace
- Developers: coding, testing & debugging
- Designers: portfolio & work-sample walkthrough
- Product: case exercise & structured reasoning
Your team makes the call
- Final technical / domain validation
- Team fit, leadership & candidate selling
- The hiring decision stays with you
See the evidence on your own roles.
We'll configure an adapter and rubric for a role you're hiring, run a shadow pilot alongside your current process, and show you the report your team would work from.
Founding-partner program · design, product & engineering hiring teams · India & remote