How Stori works
Last updated: 14 July 2026
Stori gathers, organizes, and cross-validates candidate information to help employers make more informed decisions. This page describes the system architecture in the terms a legal or procurement reviewer needs: what goes in, what the AI does with it, what it produces, and where the human sits.
The pipeline, end to end
1 · Candidate-initiated profile
A candidate builds their own Stori profile — resume, application materials, and a structured interview. The profile is candidate-owned and portable: it is the candidate's evidence, not an employer's verdict that follows them around.
2 · The structured interview
A voice-based, story-driven interview. Questions are open prompts that invite the candidate to narrate real experience; they are not trait quizzes or puzzle tests. The written transcript is the record of the interview — evaluation reads the transcript.
3 · Evidence extraction
The AI reads the transcript and identifies specific, role-relevant claims — what the person actually did, built, and was responsible for — each tied to its source quote and context (its receipt).
4 · Two separate lenses
Lens one: interview evidence, read for four meta-traits (Thinking, Discipline, Execution, Communication), scored absolutely — never zero-sum — with a stated confidence per facet, abstaining (“insufficient signal”) where evidence is thin. Lens two: a separate Big Five self-report, returned to the candidate as self-insight and shown to employers as context. The two lenses are never collapsed into one score, and the self-report never contributes to ranking, prioritization, filtering, or exclusion.
5 · Cross-validation
Where the two lenses corroborate each other, convergence strengthens the read; where they diverge, the divergence itself is surfaced for a human to weigh. Cross-validation means checking consistency across candidate-provided sources — it does not mean determining that a statement is true.
6 · Evidence ordering
For an employer's search, evidence is weighed against criteria the employer selects for their role. Ordering reflects the presence, specificity, and checkability of role-relevant evidence — “led the migration that cut costs 18% over two quarters” carries weight; “I'm a hard worker” does not. The output is a qualified pool: candidates for whom sufficient role-relevant evidence exists to justify serious consideration.
7 · Human review
Employers see the evidence, its receipts, the confidences, and the abstentions. Every interpretation can be checked against its source and overruled; overrides are captured. The employer makes the decision.
What the system is trained on — and what it is not
Stori does not train on hired / not-hired outcomes. There is no model learning patterns from a historical pile of accepted and rejected candidates, so historical selection bias has no route into the system as “ground truth.” Each candidate is assessed from their own interview and materials.
What evaluation reads — and what it never reads
- Evaluation reads the substance of what a candidate says — the clarity, coherence, and specificity of the ideas and facts in the transcript.
- It never scores speech delivery: fluency, accent, pace, hesitation, or speech patterns carry no signal, positive or negative.
- It does not perform biometric identification or verification — no facial templates and no voiceprints are generated as scoring inputs.
- Protected characteristics are not evaluation criteria, and location is used only as a job-relevant signal (proximity to a stated work location), never as a proxy for a protected class.
Where the boundaries are drawn
Two companion documents state the operating boundary precisely: what the AI decides and what it never decides. The one-sentence version: the machine is the index, not the authority.
This document describes how Stori is designed. It is not legal advice and does not determine whether any specific employer deployment is lawful. Questions from legal or procurement teams: support@onestori.com.