InterviewMate AI - Virtual Interviewer Assistant

An English-first recruiting prototype connecting batch CV screening to personalized voice and text interviews. Recruiters rank up to 50 CVs against a role, choose who to invite, and review transcript assessments with cited competency evidence, explicit missing-evidence states, and human review.

Proof points

  • Up to 50 CVs per batch
  • Cited 0–4 evidence rubric
  • 1,140 tests across 27 suites

Technologies

Next.js, TypeScript, WebRTC, OpenAI Realtime, Zustand, Firebase, Firestore, Tailwind CSS

Links

Problem

Recruiters need a consistent first-screening workflow that connects CV review to interviews while making evidence, uncertainty, and human decision ownership visible.

Approach

Built a Next.js prototype for batch CV parsing and deterministic role-based ranking, recruiter-selected invitations, candidate-specific voice and text interviews, and transcript assessments with cited competency evidence. Recent engineering work centralized provider resolution and added route-contract coverage for fallback behavior.

Results

Released a hosted recruiter workflow with scoped records, saved human review, and explicit missing-evidence states. The 2026-09-21 engineering branch passed 1,140 tests across 27 suites. The ATS and interview scores remain decision support; no hiring-validity or recruiter time-saving claim has been established.

InterviewMate AI: Recruiter Workflow and Evidence Review

Overview

InterviewMate is an English-first recruiting prototype that connects recruiter-led CV screening with candidate-specific interviews and human review. The hosted MVP is demonstrable; its scores have not been validated against hiring outcomes, recruiter time savings, or an independent model-quality benchmark.

Recruiter-led screening

  • A recruiter enters a role brief and uploads up to 50 CV PDFs.
  • The application parses CVs and applies deterministic ATS scoring to rank candidates against that role.
  • Recruiters can correct extracted identity details, inspect the ranking, choose candidates, and create invitation links. The system does not automatically reject candidates or make hiring decisions.
  • Role and interview settings are snapshotted into invitations, and recruiter records are owner-scoped.

Interview and assessment

  • Candidates complete a structured or adaptive interview through a hosted voice and text workflow using WebRTC and the OpenAI Realtime API.
  • CV context can inform interview questions. The transcript assessment uses a cited 0–4 competency rubric, makes missing evidence explicit, and gives recruiters a saved human-review workflow.
  • ATS and assessment results are review signals. The recruiter retains responsibility for interpretation and all hiring decisions.

Engineering and release evidence

  • Stack: Next.js App Router, TypeScript, WebRTC, OpenAI Realtime, Firebase Auth, Firestore, and Firebase Storage.
  • Production release (2026-09-11): 191 tests across 25 suites, TypeScript checks, production build, Node 24 CI, and a documented production smoke check passed.
  • Latest engineering branch (2026-09-21): quest/trust-and-change centralizes provider and key resolution across the three evaluation routes, makes fallback policy explicit, prevents scheduled evaluations from silently substituting a provider when fallback is disabled, and trims server keys consistently. Its resolver and route contract suites cover 214 and 735 cases; the full branch check passed 1,140 tests across 27 suites.

Evidence limits

The prototype demonstrates a recruiter-controlled screening and interview workflow. ATS ranking and AI evidence assessment are not calibrated against recruiter labels, hiring outcomes, or measured time savings. Candidate-data model training and automated hiring decisions are not enabled.