No single workflow
Screenings, care plans, scheduling, and billing don’t talk to each other. The care plan, invoice, and calendar can disagree.
PRD, design, and a live web build for pediatric therapy centers : Done using Cursor
The problem
Every conversation repeated the same breaks. Centers and parents were asking different versions of the same questions — and nobody had one trustworthy answer.
Screenings, care plans, scheduling, and billing don’t talk to each other. The care plan, invoice, and calendar can disagree.
Pricing and tax live in spreadsheets. Kids can start sessions before an invoice is paid — so revenue slips away.
Goals aren’t tied to assessment evidence. Session outcomes aren’t structured. There’s no clear trail of what worked.
No single view of who’s booked and what’s free.
Branches jump tools — no shared roster or billing.
Double-books and WhatsApp reshuffles.
“What happened last time?” lives in memory.
Parents ask weekly — answers are anecdotes.
What to practice fades after the drive home.
Research
Theraxis started from direct conversations, not a hypothesis. I talked to parents, therapists, therapy center owners, clinical admins, and center heads — primarily in Hyderabad, with the same pattern in Chennai and Pune.
I spent time with three Hyderabad centers — Daffodils Child Development Center, Ignitio Child Development Center, and Ankura Hospitals’ pediatric services — different scales, same structural gaps.
Hyderabad · Speech, OT, ABA, behavioral therapy
Hyderabad · Speech, OT, ABA, early intervention
Hyderabad, Pune + 5 cities · Pediatric specialty care
Research conversations only — not clients, not partners, not endorsements.
Hard system gates
These gates are not UI niceties — they refuse invalid states before a screen can look “empty” or permissive.
Gate 01
Refuses Care Plan generation if domain evidence is insufficient.
Gate 02
Financial confirmation / payment gates session start and resource allocation.
Gate 03
Availability grid disables conflict slots natively — no double-book path.
Gate 04
RLS policies enforce tenant/role isolation at the DB layer, not via UI filtering.
Gate 05
Clinical vs financial domains separated by role — billing noise can’t bleed into care.
Personas
Staff need continuity between sessions. Families need a clear answer to “is my child improving?”
Persona 01
“What did we work on last session — and what should I do today?”
Opens every day
Problem
Needs
Persona 02
“Is my child getting better — and what should I do at home?”
Opens every day
Problem
Needs
What I did
End-to-end ownership — same person from field notes to a running app in Cursor.
Centers, therapists, parents — Hyderabad first, then the same pattern elsewhere.
Roles, five hard gates, workflows, access — enforced in the database, not the UI.
Staff web for roster, screening, assessments, billing, and admin overview.
Working multi-tenant app (Cursor). Tested with clinicians. Parent mobile next.
Product
Screens from the live staff web build — not mockups.
01 · Center pulse
Students, revenue, staff — one tenant view.
02 · Roster
Status on the board, then the full record.
03 · Evidence
Age-band checklists → domain bars — not anecdotes.
04 · Clinical + billing
Age-filtered tools — approve and price before work continues.
Status
Working build. PRD, gates, and architecture are implemented — screenshots above are from the live app, not mockups.
Tested directly with practicing doctors, who responded well — they said it addressed real clinical and operational gaps, and that they’d expect to pay more for something that solves those problems than for a typical practice-management tool.
Not a center’s system of record yet. Parent mobile still under construction. Next phase is rollout validation.
Product requirements
Full PRD Goals, roles, gates, workflows, access matrix, architectureThis PRD represents the work honestly — systems rules, not just screens. The application is built and running; screenshots on this page are from the live build. I built it end-to-end with Cursor.
Role: Product & Systems Design — built end-to-end
Multi-currency billing and non-INR pricing are out of scope — the spec targets Indian therapy centers on GST-based pricing. Insurance claims processing, non-pediatric therapy types, were deferred for v1. The Parent experience is a separate mobile app under construction — not part of the staff web product.
| Role | Data Scope | Core Capability | Cannot Do |
|---|---|---|---|
| Receptionist | Own center only | Intake, appointment booking, session reschedule/cancel | Clinical records, billing, care plans |
| Center Manager | Own center only | Admissions, enrollment, invoicing, session scheduling | Clinical notes, care plan approval |
| Senior Therapist | Center + supervisees | Care plan generation/approval, pre-session tasks, session review | Billing, staff management |
| Therapist | Assigned children only | Assessments, care plan drafts, session conduct, MCQ submission | Billing, scheduling, unapproved plans |
| Tenant Admin | All centers | Everything across all centers | Nothing — full access |
| Parent | Own child only | Progress view, session calendar, home carryover | Clinical data, billing, staff info |
"I need to register a new child and get them booked in under five minutes."
"I need to turn a care plan into a running therapy schedule the same day the family says yes."
"I approve care plans. I need the full picture before I can sign off."
"I open a session, work through the checklist, write my notes, and submit."
"I need to see across every center I run, not just one — and I need to be the fallback when something's stuck."
"I just want to know my child is improving, and what I should be doing at home."
The system refuses to generate a plan if domain evidence is insufficient.
Centers can't commit clinical resources before financial confirmation. Enforced at the API layer; Center Manager can override with a recorded reason.
A slot is either selectable or it isn't — no warnings to dismiss, no overrides to design.
Role permissions are Supabase RLS policies — a direct API query still only returns a therapist's assigned children.
Therapists never see prices; Center Managers never touch session notes or care plans.
Completed notes and MCQ data automatically feed the evidence engine for the next cycle.
| Capability | Recept. | Ctr Mgr | Sr. Ther. | Therapist | Tenant Admin | Parent |
|---|---|---|---|---|---|---|
| Create students | — | — | — | — | ||
| Run assessments | — | — | — | — | ||
| Generate care plans | — | — | — | |||
| Approve care plans | — | — | — | — | ||
| Create admissions & invoices | — | — | — | — | ||
| Schedule sessions | — | — | — | — | ||
| Conduct sessions + MCQs | — | — | — | — | ||
| View billing & invoices | — | — | — | — | ||
| Manage staff & centers | — | — | — | — | — | |
| View child progress | — | — |
The sharpest tension in this project was between the AI-assisted care plan draft and clinical governance: an evidence engine that synthesizes 14 assessment domains into a draft plan is only useful if a Senior Therapist can trust it, and trust required making every gap and low-confidence domain visible rather than letting the draft look more finished than the evidence supported. That's why the draft is never auto-published — a human reviews and edits all six sections before anything moves to Approved.
Enforcing the five hard gates at the database layer rather than in the UI changed how the rest of the spec was written: a rule that only lives in a screen is a suggestion, not a guarantee. Writing gates as Row-Level Security policies forced every requirement to be stated precisely enough for a database to enforce it, not just precisely enough for a designer to explain it.
With more time, the next area to pressure-test is the override path on payment gating — a Center Manager can unlock a session before payment clears, with a recorded reason. That's the one place the system trades a hard rule for judgment, and it deserves the same scrutiny the other four gates got before this goes further.
Confidentiality note: This project is under NDA. Design is complete and moving into development.
Interface redrawn and figures synthesized for case study presentation.
The core architectural logic and UX constraints remain true to production.