feelxs
a non-verbal and verbal assessment for hiring the people we trust to care for others.
Most hiring screens for clinical and care-facing roles still rely on the same instruments that have always failed quietly: a CV, a structured interview, a personality questionnaire taken on a laptop. feelxs is an attempt to look, really look, at how a candidate shows up when asked to sit with another person's difficulty. It is built for hospitals, clinics, schools, and the small number of organisations whose work depends on the relational quality of the people doing it.
The teams who do the most relational work (children's hospitals, mental-health services, schools) have the least time to do hiring carefully. The cost of a poor hire is paid for slowly, in turnover, in patient experience, and occasionally in harm. The instruments meant to help with the selection, assessment centres, personality inventories, reference calls, were not designed for the question that actually matters.
feelxs was built around a more honest question. When a candidate is asked to address a patient or a child in real difficulty, what do they do with their face, their voice, their words, and the silence between them?
Candidates record short video responses to a small set of scenario prompts. The platform transcribes the audio (Deepgram), extracts eight acoustic features per response (parselmouth: pitch, intensity, jitter, shimmer, and four others), samples frames at half-second intervals, and passes the result to a vision-capable model with the transcript and acoustic features as context.
The output is five layered analyses: a twelve-dimension feelxs score, a five-dimension non-verbal ENACT report with cited frame timestamps, a per-scenario breakdown, an OCEAN/Big-Five estimate, and a supervision-plan recommendation. The whole pipeline runs automatically; the admin reviews the report, not the raw video.
The first production cohort, nine candidates across our pilot organisations, produced reports that hiring managers described as the first time they had seen a candidate before meeting them. The non-verbal layer surfaces what interviews are designed to hide: hesitancy where there should be presence, fluency where there should be care.
Early outcomes data, feelxs scores correlated against later PHQ-9 trajectories of the people the candidates went on to support, is being typeset at stats.feelxs.org.
feelxs.org is live; the pilot at a leading US children’s hospital is in its second cohort. We are talking to hospitals, school systems, and youth-services organisations about further pilots and to the supervisory bodies who would eventually need to certify the use of an instrument like this.