Built for the workforce Lorin's dissertation centers: the psychiatrists, counselors, social workers, and direct-care staff whose own burnout stays invisible. Self-assessment, AI support that never sleeps, and human oversight so AI never carries care alone.
Your load eased since Thursday. Two recharge habits are holding.
Today
Over the past week, how true has each felt?
Drawn from the ProQOL (compassion satisfaction, burnout, secondary traumatic stress) plus a brief PHQ-2 / GAD-2 mood scan.
I feel invigorated after working with the people I help.
Compassion satisfaction is still a strength. Secondary traumatic stress ticked up this week.
A licensed clinician reviews any elevated result within 24 hours. Your employer only ever sees anonymous, team-level trends.
Recommended for this week
Completed
Talk to someone
This is the prototype exactly as it stood before Lorin's UI/UX review, kept so the two can be compared side by side. Switch to Worker app v2 to see what her feedback changed.
What v1 got wrong
v1 led with the model instead of the person: a JD-R read, demands and resources, three raw subscale scores, a gamified streak. Every number was right. None of it told a burned-out counselor what to do in the next five minutes.
A JD-R gauge and a Today list. v2 opens with "How am I doing?" and three things you can actually do.
v2 calls it AI Support, labels it, and offers four ways in before the chat opens.
Gamification punishes the week someone was too flattened to open the app. v2 says "You practiced 3 skills this week."
Burnout: Elevated, and nothing else. v2 adds what it means, what changed, what's helping, and what to do next.
v1's check-in header said "never shared with your name", then a named clinician reviewed her named scores. v2 promises only what is true: your employer never sees this.
Your recovery is keeping up with what work takes out of you.
Sleep dipped two nights, from your wearable.
Human support
Peers are matched outside your own team, trained, and bound by the same confidentiality as your clinician. You choose whether to use one, and they see only what you send them.
Over the past week, how true has each felt?
Six taps, about 2 minutes. No wrong answers, and nothing leaves this phone with your name on it.
I feel invigorated after working with the people I help.
AI identifies patterns. A licensed clinician reviews elevated concerns.
What you can do next
A licensed clinician reviews any elevated result within 24 hours. Your employer only ever sees anonymous, team-level trends.
By what's weighing on you
Talk to someone
This is the one place that holds you. AI notices the patterns, and a licensed clinician stays in the loop so you are never handed to a machine alone.
You should not have to guess at this, and you should not have to take our word for it either.
Pick anything that fits. You can change this whenever you want.
Six taps, about 2 minutes. From there you get a recommendation right away, AI Support whenever you want it, and a clinician if things climb. Weekly check-ins after that build the pattern over time.
Home adapts to what the check-ins say. At high distress the screen collapses to four ways out, because reading a dashboard is the last thing anyone needs then.
The synergy, made literal
Lorin's thesis is that neither works alone: AI without oversight dehumanizes, human-only support doesn't scale to a burned-out workforce. Every teal touch is the machine; every apricot touch is a person.
ProQOL, PHQ-2 and GAD-2 surfaced as gentle weekly check-ins, not clinical intake forms.
AI notices continuously, a trained peer responds first, a licensed clinician takes over when it climbs. That is the Scott three-tiered second-victim model, not a feature list.
Five-minute, evidence-based lessons on burnout, boundaries and secondary trauma.
Passive sleep and stress data flag a rough stretch before you'd name it yourself.
Average demands-vs-resources score, weekly. Higher is healthier.
Where the team sits this week. No individual is identifiable.
The app measures the load. Only you can change it. Ranked by projected effect on team balance, with what the check-ins point at.
Six of the eight staff in the Elevated band sit in crisis intake. Their recovery scores fall on intake days and recover on non-intake days. Basis: 3-week trend, intake rota cross-referenced with check-in timing. n=8, above the display floor.
Back-to-back intakes with no buffer is the single most common pattern behind the same-day drops. This is a scheduling change, not a wellness offer. Basis: same-day balance deltas on days with 3+ consecutive intakes.
Recovery scores drop on days with the heaviest note-writing load. Staff are finishing notes on their own time. Basis: balance dips correlate with note-completion volume, not session count.
Two trained peers currently cover 142 staff, and peer contact is the support your people actually accept. Coverage, not willingness, is the bottleneck. Basis: peer contacts requested vs fulfilled this month. Peer support has the strongest short-term evidence of any second-victim intervention.
Two teams moved to four longer days. Balance is up 11 points in that group since the change. Widen it or stop it on the next read. Basis: pre/post comparison, 22 staff, 6 weeks.
The AI raises its hand. You decide what it means. It books nothing and diagnoses nobody.
Everything the AI can hand you, and the wall it cannot reach past.
The 2024 dissertation predates the first real trials of generative-AI therapy. Here is the refreshed evidence base, verified through OpenEvidence, and the three design calls it points to.
What changes for the app
Every 2025-26 review lands on supervised hybrid stepped-care. Illinois banned AI therapy without a licensed clinician; the APA says don't let generative AI deliver therapy. The clinician hand-off isn't a nice-to-have anymore, it's the requirement your design already meets.
The 2026 JAMA Psychiatry attrition study (79 RCTs) found reminders and human contact lower dropout, while gamification is tied to higher dropout. Lead with reminders, human contact, personalization, and symptom monitoring instead.
Effects are modest and fade past 3 months, so the bar for a medical claim is real. Launch as a wellness / organizational tool, gather real-world evidence, then pursue FDA De Novo and a German DiGA listing (insurance-reimbursed).
Clinicians whose care goes wrong, or gets attacked in public, are what the patient-safety literature calls second victims. The standard response is the Scott three-tiered model: everyone trained to notice it, trained peer supporters as first responders, then expedited referral to professional care. This app maps onto those tiers directly: AI notices, a peer responds, a licensed clinician takes over when it climbs.Scott three-tiered model · Jt Comm J Qual Patient Saf 2024
The honest read on the evidence: a 2025 systematic review of 15 studies, GRADE-assessed, found peer support programmes give consistent short-term benefit on distress and perceived isolation, but evidence for long-term burnout reduction and retention was mixed and of low certainty. Structured CBT and mindfulness looked more promising long-term and remain underexplored. So the peer tier is the best-supported thing here for the acute window, and nobody should claim it fixes burnout.Syst. review, J Healthc Leadersh 2025
Two design consequences. First, the preferred source of support in this literature is plainly stated as "a respected peer", not a manager and not a stranger, which is why the peer is a real colleague from another team rather than a call centre.Nurs Clin North Am 2024 Second, a peer inside your own workplace is a confidentiality problem, so peers are matched outside the worker's own team and see only what the worker sends them.
Employer-provided mental health apps routinely hand the employer more than they admit. Mozilla Foundation documented that Headspace for Work's admin dashboard exposes each employee's first name, last name, registration date and last-use date, and named the obvious misuse: an employer reading engagement as productivity and allocating bonuses by department.Mozilla Foundation 2022
That is the context every worker brings to an app their employer paid for. Thirty-seven percent of US workers already link workplace monitoring to worse mental health, and 43% say monitoring makes them feel their employer distrusts them.APA · Staffing Industry Analysts Meanwhile traditional EAP utilisation sits near 4%, with confidentiality named as a top barrier to use.EAP utilisation reviews 2026
So the wall is not compliance decoration. It is the conversion mechanism, and it only works if it is demonstrated rather than asserted, which is why onboarding hands the worker the employer's actual screen before asking them a single question.
Full write-up with effect sizes and APA-7 references is in the companion literature-update addendum. Therabot (NEJM AI, 2025) is cited from the primary source, it was not indexed in OpenEvidence at the time of writing.
Working a psychiatric center "makes you a very angry person." This room takes that seriously: hit something that can take it, then learn the regulation skill while your nervous system is actually listening. AI Support builds tonight's room from your check-in.
Tonight's room is built from your check-in, three crisis intakes, tension running high.
Step in when you're ready. Nothing in here needs you to be okay yet.
You named it, released it, and chose your next move. That's the whole skill.