Recrewty
A specialised hiring chatbot, built on Fabrile and validated in peer-reviewed research — personality inference that holds up under academic scrutiny.
The Challenge
Recrewty, a Montenegrin HR-tech startup, wanted to answer a question most recruitment AI never tests honestly: can an AI chatbot infer personality for hiring as reliably as an established psychometric test — and resist the social desirability bias that distorts self-reports in high-stakes selection? Answering it credibly meant the chatbot couldn't be a demo. It had to be a research instrument: consistent, measurable, and rigorous enough to stand up in a peer-reviewed study. It also had to work in Montenegrin and Serbian — a language pairing unsupported by the chatbot platforms used in earlier research. Recrewty and the study's authors engaged Ingram to build it.
Our Approach
Build the chatbot as a research instrument on Fabrile
Ingram built and customised the chatbot on Fabrile, our no-code agent platform — chosen because it could be tailored to run the assessment in Montenegrin and Serbian. It ran on OpenAI's GPT-4 API, with no language-specific training phase, relying on careful design rather than a large training corpus.
Implement a novel one-question-per-facet design
Rather than scoring a whole transcript at once, the chatbot asked one open-ended question per facet of the Big Five and scored each individually — a methodological advance that allowed the study to evaluate the chatbot's psychometric properties at a granular level.
Support a rigorous, real-world study
The chatbot underpinned a quasi-experimental study of 159 professionals across Serbia and Montenegro, using propensity score matching to compare a control group against real bank-hiring candidates — measuring its inferences against a traditional 50-item Big Five questionnaire.
Collaborate through to peer-reviewed publication
Ingram developed the chatbot alongside the authors and provided the Fabrile platform to the research free of charge. The work was accepted to Frontiers in Psychology, co-authored with the London School of Economics' Department of Psychological and Behavioural Science.
What We Deployed
- •A specialised personality-inference chatbot, built and customised on Fabrile
- •A GPT-4-based assessment localised for Montenegrin and Serbian
- •A novel one-question-per-facet design for facet-level psychometric analysis
- •The instrument behind a 159-participant, quasi-experimental study
Outcomes
- •The chatbot showed good structural, substantive, and convergent validity for Extraversion and Conscientiousness
- •AI-inferred scores were found robustly less susceptible to social desirability bias than traditional psychometric tests
- •Honest findings, too: weaker validity on other traits and lower predictive validity than traditional tests — the kind of result only rigorous research surfaces
- •Accepted to Frontiers in Psychology (2025) — peer-reviewed validation of Recrewty's research-led approach
“A sincere and big thanks to the Ingram Technologies team, who did an amazing job and helped with the technical deployment of the specialised chatbot — which showed results on par with leading models.”
Have a similar deployment in mind?
Tell us what you're building. We'll tell you what it takes to deploy it — reliably, securely, and at scale.