← All case studies TheraLoop
TheraLoop · Clinical Intelligence · Harbor H001
AI-powered clinical intelligence platform for outpatient physical therapy. Live in production.
Status
Live
Intake
AI via SMS
Documentation
Minutes
PHI exposure
ZERO
01 · CONTEXT
Who. What scale. What was at stake.
An outpatient physical therapy practice needed a system that handled the full patient lifecycle without requiring the clinical team to operate six disconnected tools. Intake was paper. Documentation took an hour after clinic. Home exercise programs were printed handouts. Patient engagement between visits was nonexistent. Every task that did not require clinical judgment was still consuming clinical time.
02 · CONSTRAINT
The architectural problem.
MILO had to understand a patient's clinical profile without ever seeing the raw clinical record directly. Weight-bearing status, post-op constraints, documented precautions: all had to inform what MILO suggested without PHI entering the intelligence layer. The second constraint: every clinic that deploys TheraLoop gets its own sealed environment. One clinic's data, branding, staff, and configuration cannot commingle with any other. The architecture had to support multi-clinic scale from day one.
03 · DECISION POINTS
Three decisions that shaped everything downstream.
Decision 01
AI handles first contact, humans handle judgment
Every patient-facing channel runs through AI first. Intake collection, scheduling requests, routine questions, and HEP delivery are handled autonomously by MILO. Escalation to a human only when clinical judgment is required. Deterministic, audit-logged. Not a chatbot answering FAQ. An operational layer that qualifies, summarizes, and routes.
Decision 02
Structured intake before the first visit
MILO runs conversational intake via SMS before the patient arrives. Region, mechanism, severity, goals, and functional outcome scores collected in plain language and structured for the clinical record. Evaluation form pre-filled from intake. Collapsible sections, sticky navigation, eight auto-populated goals by region. Documentation that takes minutes, not the hour after clinic.
Decision 03
HEP intelligence tied to the clinical record
When a PT assigns a home exercise program, the system understands the patient's clinical profile. Weight-bearing status and post-op constraints filter what MILO will suggest. MILO will not recommend load-bearing exercises for a patient flagged as non-weight-bearing.
04 · SYSTEM
What was built.
Patient intake: MILO-driven conversational intake via SMS. Chip-based inputs. Structured clinical data lands in the eval before the first visit. No clipboards. No transcription.
Clinical workspace: SOAP documentation, evaluation forms, HEP assignment, scheduling, and care coordination in one session. Sections default collapsed with multi-open support. Provider-signed, timestamped documentation.
Patient portal: web and iOS. Care plan, home exercise program, visit schedule, and care team messaging through MILO or directly through the portal.
MILO: available 24/7. Handles routine questions, delivers exercise programs, logs patient progress, escalates to the care team when it matters. Refers to patients as "the patient" in clinical contexts because it operates on de-identified context.
Admin tooling: PT Directory, Patient Directory, Harbor Settings, Billing Export. Admin navigation gated by RBAC. Clinical leads see the full roster. Clinicians see their assigned patients only.
Harbor architecture: Harbor H001 is live in production. Dedicated data stores, isolated network, customer-managed keys. buildTheme applies the clinic's visual identity with zero code changes.
05 · OUTCOMES
All metrics from production systems.
Harbor H001 live.
AI-driven intake collecting structured clinical data before the first visit.
Clinical documentation reduced from post-clinic hours to minutes during the session.
Home exercise programs delivered through MILO with clinical constraint filtering.
Patient engagement between visits handled autonomously.
Zero PHI exposure in the intelligence layer, by architecture.
Admin tooling live: PT Directory, Patient Directory, Harbor Settings, Billing Export.
Platform architecture validated for multi-clinic expansion without code changes.
06 · DISCUSS FURTHER
The architecture above is public. What follows is a conversation.
MILO agent specification, intake flow design, eval form architecture, HEP constraint logic, Harbor provisioning process, RBAC permission model, and the specific clinical workflow decisions are shared in a 30-minute conversation with healthcare operators evaluating similar clinical intelligence work.
Start a conversation →
← mastOS All case studies Healthcare AI →