NeuroStaff — AI-HR SaaS for automated first-line hiring

Context

In high-volume hiring the first pass — reading resumes, running screening calls, forming a first impression — consumes the most recruiter time and adds the least signal. It is repetitive, subjective, and impossible to scale by simply hiring more recruiters.

NeuroStaff automates that first pass end to end and hands humans a ranked shortlist, so recruiter time is spent on the actual decision rather than on triage.

Approach

The funnel is decomposed into four automatable stages. Resumes are screened against the role profile before any human touches them. A voice agent then conducts a structured screening interview — the same questions in the same order for every candidate — transcribed by STT. An LLM scores soft skills from the transcript and assembles a comparative rating.

Because every candidate passes through an identical structured pipeline, they are compared on the same axes — not on whoever the recruiter happened to speak with last, or in what mood.

Architecture

Six components form the system: an hh.ru integration for sourcing, the voice agent, the STT layer, the LLM scoring engine, a candidate database as the system of record, and an HR dashboard with reporting on top. The pipeline is staged, so each step leaves an auditable artifact — the raw resume match, the interview transcript, the per-axis scores.

The candidate database makes the funnel cumulative: every screened candidate, every transcript and every score is retained, so the shortlist for a new opening can draw on previously assessed people rather than starting from zero.

Key engineering details

The critical design constraint is explainability of ranking. A shortlist position is not a black-box number: it decomposes into per-axis scores, each traceable to transcript fragments, so a recruiter can challenge and override any individual judgement. That traceability is also the foundation for the planned bias and fairness checks on the ranking.

Structured interviews are what make LLM scoring reliable at scale — a fixed question protocol bounds the variance of the input, so score differences reflect candidates, not conversations.

Outcome and what shipped

The product architecture is complete and in development, ready to be assembled into an industry-specific MVP: four funnel stages automated, six components specified, sourcing from two channels, and results delivered on three surfaces. The target use cases are primary selection, mass hiring, AI-conducted interviews and comparative candidate evaluation.

Next steps are defined: assemble the first vertical MVP, calibrate scoring against real hiring outcomes, and add fairness checks to the ranking.

What we built

• Resume screening

  Candidates screened against the role before a human spends time on them.

• Voice interviews

  A voice agent runs structured screening interviews, transcribed by STT.

• Soft-skill scoring

  An LLM scores soft skills and builds a comparative rating.

• HR dashboard & reports

  A candidate database, dashboard and reporting make the funnel measurable.