@inproceedings{kappaz-etal-2026-caliber,
title = "{CALIBER}: Calibrated Assessment via {LLM} for Interview Behavioral Evaluation and Rating",
author = "Kappaz, Katherine and
Alavo, Raymond and
Calderon, Sofia and
Abbasi, Ahmed",
editor = "Agrawal, Sweta and
Abad, Alberto",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-demos.45/",
pages = "466--475",
ISBN = "979-8-89176-461-3",
abstract = "We present **CALIBER**, an interactive dashboard for calibrated, LLM-assisted behavioral interview evaluation. The system is built around three core design principles. First, a structured **Annotate{--}Rubric{--}Rate (ARR)** workflow sequences each evaluation into span annotation, rubric consultation, and scoring, decomposing a holistic judgment into auditable steps. Second, LLM-generated STAR span highlights serve as **attention scaffolds** rather than scores: the model surfaces structural cues in the transcript while all final ratings remain the annotator{'}s responsibility. Third, a 21-step guided training tour with Behaviorally Anchored Rating Scales (BARS) brings every evaluator to a shared baseline before live annotation begins. On the backend, a shrinking-pool randomizer with atomic DynamoDB counters guarantees uniform item coverage across crowdworkers, and a rich telemetry stream captures click, mouse-movement, and timing data alongside every annotation. A user study (N=24) confirms that the ARR design achieves its goals: both AI-friendly and AI-averse annotators complete tasks with less effort under AI assistance, and alignment with a held-out LLM reference improves for all groups. We release CALIBER as an open-source toolkit together with a live demo and code."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kappaz-etal-2026-caliber">
<titleInfo>
<title>CALIBER: Calibrated Assessment via LLM for Interview Behavioral Evaluation and Rating</title>
</titleInfo>
<name type="personal">
<namePart type="given">Katherine</namePart>
<namePart type="family">Kappaz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Raymond</namePart>
<namePart type="family">Alavo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sofia</namePart>
<namePart type="family">Calderon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ahmed</namePart>
<namePart type="family">Abbasi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sweta</namePart>
<namePart type="family">Agrawal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alberto</namePart>
<namePart type="family">Abad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-461-3</identifier>
</relatedItem>
<abstract>We present **CALIBER**, an interactive dashboard for calibrated, LLM-assisted behavioral interview evaluation. The system is built around three core design principles. First, a structured **Annotate–Rubric–Rate (ARR)** workflow sequences each evaluation into span annotation, rubric consultation, and scoring, decomposing a holistic judgment into auditable steps. Second, LLM-generated STAR span highlights serve as **attention scaffolds** rather than scores: the model surfaces structural cues in the transcript while all final ratings remain the annotator’s responsibility. Third, a 21-step guided training tour with Behaviorally Anchored Rating Scales (BARS) brings every evaluator to a shared baseline before live annotation begins. On the backend, a shrinking-pool randomizer with atomic DynamoDB counters guarantees uniform item coverage across crowdworkers, and a rich telemetry stream captures click, mouse-movement, and timing data alongside every annotation. A user study (N=24) confirms that the ARR design achieves its goals: both AI-friendly and AI-averse annotators complete tasks with less effort under AI assistance, and alignment with a held-out LLM reference improves for all groups. We release CALIBER as an open-source toolkit together with a live demo and code.</abstract>
<identifier type="citekey">kappaz-etal-2026-caliber</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-demos.45/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>466</start>
<end>475</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T CALIBER: Calibrated Assessment via LLM for Interview Behavioral Evaluation and Rating
%A Kappaz, Katherine
%A Alavo, Raymond
%A Calderon, Sofia
%A Abbasi, Ahmed
%Y Agrawal, Sweta
%Y Abad, Alberto
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-461-3
%F kappaz-etal-2026-caliber
%X We present **CALIBER**, an interactive dashboard for calibrated, LLM-assisted behavioral interview evaluation. The system is built around three core design principles. First, a structured **Annotate–Rubric–Rate (ARR)** workflow sequences each evaluation into span annotation, rubric consultation, and scoring, decomposing a holistic judgment into auditable steps. Second, LLM-generated STAR span highlights serve as **attention scaffolds** rather than scores: the model surfaces structural cues in the transcript while all final ratings remain the annotator’s responsibility. Third, a 21-step guided training tour with Behaviorally Anchored Rating Scales (BARS) brings every evaluator to a shared baseline before live annotation begins. On the backend, a shrinking-pool randomizer with atomic DynamoDB counters guarantees uniform item coverage across crowdworkers, and a rich telemetry stream captures click, mouse-movement, and timing data alongside every annotation. A user study (N=24) confirms that the ARR design achieves its goals: both AI-friendly and AI-averse annotators complete tasks with less effort under AI assistance, and alignment with a held-out LLM reference improves for all groups. We release CALIBER as an open-source toolkit together with a live demo and code.
%U https://aclanthology.org/2026.emnlp-demos.45/
%P 466-475
Markdown (Informal)
[CALIBER: Calibrated Assessment via LLM for Interview Behavioral Evaluation and Rating](https://aclanthology.org/2026.emnlp-demos.45/) (Kappaz et al., EMNLP 2026)
ACL