SURGE: Building an AI-Powered Digital Wellbeing & Healthcare Platform

This is the first system a user touches, and it collects structured answers across profession, sleep, social connection, physical activity, stress triggers, personality, and prior mental-health history rather than a generic three-field signup form. The core decision was to treat onboarding as a data-collection instrument, not a registration step; every field maps directly to something a later system reads, whether that's the AI's opening tone or which professionals get surfaced first. The constraint was length versus completion: too short and the AI has nothing to work with, too long and users abandon before finishing, so the questionnaire is broken into defined categories instead of open free-text. This prevents the common failure mode in this category: an AI that greets every user identically regardless of what it could already know about them.

The conversational AI doesn't have one static interface; its gradient and color treatment shift across states, calm, energetic, comforting, focused, and others, tied to the emotional context of the interaction. The interaction structure stays identical across every state; only its visual expression changes, a deliberate choice over building separate interfaces per mood, since a separate build per state would have multiplied engineering and QA effort for no functional gain. The tradeoff was consistency versus expressiveness: users need to recognize the same assistant every time, but the product also needed color and tone to communicate emotional register the way a human conversation partner would. This prevents the flat, clinical feel that pure-text mental health chat interfaces tend to have.

Voice input sits alongside text as a first-class input method, with microphone-based controls and call-related interaction extending the model from typed conversation toward something closer to a phone call. Voice was prioritized because typing is itself a barrier during high-stress moments; someone mid-panic is less likely to compose a sentence than to speak one, and a text-only assistant silently filters out exactly the users who need it most urgently. The constraint was making voice and text state-compatible, so a conversation started by typing and continued by voice doesn't lose context partway through.

Rather than a single daily mood score, check-ins log three separate dimensions, Energy, Mood, and Stress, each with its own history rather than a blended composite. Keeping the dimensions separate was the deliberate choice over a single score, because a blended number hides which specific dimension is driving a bad day, and that distinction matters for both the AI's response and, downstream, for what a professional sees if the user later shares their history. The interface deliberately uses plain emotional language, happy, tense, anxious, irritated, instead of clinical terminology, designed around sustaining daily engagement rather than requiring a full assessment every time.

Professional discovery is built around expertise, experience, ratings, availability, and full profile information surfaced before any booking action, rather than a bare name-and-schedule listing. The core decision was to treat discovery as a decision-support problem, not a search problem; a directory that returns forty names sorted by availability doesn't actually help someone choose, it just moves the hard decision downstream. This was designed around the constraint that trust in a healthcare context can't be manufactured through design alone; it has to come from real credentialing data being genuinely visible before a payment decision.

Payment, pricing, subtotal, GST, card and UPI methods, happens inside the same flow as booking rather than handing the user to an external payment screen. Keeping payment in-flow was chosen specifically to avoid the drop-off point that happens whenever a user is asked to leave a booking screen to pay somewhere else, a known failure point in telehealth funnels generally. The tradeoff was implementation complexity: supporting multiple payment methods and itemized pricing inside one continuous screen is harder than redirecting to a processor's hosted page, accepted because it removes the single highest-friction step in the booking journey.

Therapists and physiotherapists operate inside a dedicated, enterprise-grade environment covering upcoming schedules, pending approvals, patient discussions, calendars, completed consultations, revenue, and withdrawals, not a stripped-down calendar bolted onto the consumer app. This was built as a genuine second product experience rather than an admin panel, because a provider who has to manage patients and payments through a limited interface will simply use a different tool for the parts that don't work, fragmenting the same problem the patient side was built to solve.

Providers upload supporting documents, degrees, certificates, clinic or hospital information, and move through defined review and approval states before appearing in patient-facing discovery. This exists specifically to make the trust signals shown to patients real rather than self-reported, since a discovery layer built around credentials and ratings only works if those credentials have actually been checked.