A privacy-first platform that turns tiny daily signals into timely, human-sized support.
The problem
Between visits, signal is sparse and noisy. Most trackers collect data but don’t convert it into timely actions or useful clinical context. Kay closes that gap with a safety-bounded, agentic system that turns brief daily inputs into earlier course-corrections and concise provider snapshots.
Kay's Core Loop
One simple idea runs the whole system: Kay takes in tiny daily signals, makes sense of them, offers one safe next step, then gets a bit smarter from what actually helped—always within your privacy and clinical guardrails.
1
Observe
What happens: Kay collects small, low-burden inputs—your ~60-second AM/PM check-ins, optional notes, and simple engagement signals. (You can also opt in to basic context like sleep duration from Apple Health.)
Why it matters: Good support starts with just enough signal, not an exhausting questionnaire.
Controls: Private by default. HIPAA-compliant safeguards. You choose what to share and with whom.
2
Reason
What happens: Multi-signal models look for short-horizon change—the early “drift” that often precedes tougher days. Kay produces a plain-English risk posture (no diagnoses) and a short “because” summary.
Why it matters: Spotting change across a few days lets you course-correct sooner, not after a crisis.
Controls: Conservative defaults; if confidence is low, Kay stays gentle and avoids over-steering
3
Act
What happens: A constrained decision policy picks one human-sized step from a vetted library grounded in CBT, DBT, Mindfulness, Motivational Interviewing, and Emotion Regulation—for example, a 2-minute grounding, a quick reframe, or a simple connection prompt.
Why it matters: One clear step beats a long to-do list; it’s actionable now.
Controls: Clinic/clinician preferences (emphasize/avoid lists, tone, sensitivities) are honored. In crisis contexts, guidance is template-locked to emergency resources (e.g., 988 in the U.S.).
4
Learn
What happens: Thumbs-up/down, which tools you used, and provider notes feed back in—so Kay surfaces what actually helps you next time. Preferences are versioned with an audit trail.
Why it matters: Support gets more relevant without getting riskier or opaque.
Controls: Safety and privacy guardrails are never overruled by learning signals
Why this loop works
- Tiny inputs → usable signal (you keep your energy)
- Early patterning → earlier support (fewer “how did we get here?” moments)
- One step → real follow-through (less overwhelm)
- Learning → fit improves over time (without losing safety or control)
Bottom line: The loop keeps Kay simple to use, clinically respectful, and HIPAA-compliant—so support shows up before things spiral, and documentation stays clean when you choose to share it.
The Techy Stuff
Data in (privacy-first)
Minimal structured inputs, optional notes, optional integrations—all user-controlled. No ad tech, no data sales. HIPAA-compliant operations; BAAs available.
NLP & understanding
Modern NLP embeddings summarize optional notes for sentiment/intent while a dedicated safety classifier routes crisis language to template-locked guidance (e.g., 988 in the U.S.). No diagnosing; plain-language “because” statements explain suggestions.
Personalization with guardrails
Clinic/clinician preferences (emphasize/avoid lists, tone, sensitivities) compile into policy constraints Kay honors across patients. Patient ratings and “My Toolbox” usage inform what’s likely to help—without overriding safety.
What you get:
- Pre-session snapshot: past-week trend, engagement highlights, and what helped—scannable in under a minute.
- Monthly rollup: time-stamped summary exported to PDF/CSV for your EHR.
Safety, fairness & quality
- Safety-first suggestions from a vetted library of low-risk micro-steps.
- Calibrated predictions validated offline and through ongoing real-world use.
- Bias checks and fairness reviews across cohorts; rapid rollback if something regresses.
You lead. Kay adapts.
Set emphasis/avoid lists at the school or technique level—CBT, DBT, Mindfulness, Motivational Interviewing, Emotion Regulation—so Kay reflects your program’s style between visits. Examples: emphasize CBT reframes in early work, de-emphasize journaling for specific patients, foreground DBT distress-tolerance during high-stress periods, keep tone MI-aligned. Preferences are versioned with an audit trail and honored across all linked patients.
(Compliance note: You retain clinical judgment. Kay’s guidance is supportive and educational, not treatment.)
