Mental Health · Digital Care Navigation
HIPAA Compliant Mental Care Agent
SaaStoAgent built a production-grade agentic care navigator for Psyter, guiding patients from intake to provider matching, booking, payments, and post-booking follow-ups, with safety and clinical governance at every step.
The Problem
Mental health care navigation is complex. Patients seeking therapy face a fragmented journey: searching for the right provider, understanding insurance coverage, evaluating therapeutic fit, scheduling appointments, and managing payments, all while in a vulnerable state.
Patient Challenges
- Finding the right therapist involves navigating multiple directories with limited context
- Therapeutic fit is critical but hard to evaluate before the first session
- Booking, rescheduling, and payment workflows are disconnected
- Patients in distress need support that is immediate, not delayed by manual triage
Platform Challenges
- Provider matching requires structured reasoning across specialization, availability, and patient preference
- Safety escalation must be systematic, not ad-hoc
- Regulatory and clinical governance requirements demand full auditability
- A single chatbot cannot handle the breadth and depth of this workflow
What SaaStoAgent Built
SaaStoAgent designed and built Psyter's Patient Agent as a multi-agent agentic system, not a chatbot. Each stage of the patient journey is handled by a purpose-built agent with defined inputs, outputs, and safety boundaries. The system orchestrates these agents into a controlled, auditable workflow that supports patients conversationally while enforcing clinical governance at every decision point.
This Is an Agentic System, Not a Chatbot
- Multi-agent orchestration: each agent is scoped to a specific domain (intake, matching, booking, safety, payments)
- Controlled workflows: state transitions are explicit and logged, not free-form LLM output
- Safety-first design: the Safety Agent runs in parallel and can interrupt or redirect any workflow
- Explainable decisions: every provider recommendation, safety triage, and booking confirmation is traceable
Multi-Agent Architecture
The Psyter Patient Agent is composed of specialized agents, each responsible for a distinct part of the care navigation workflow:
Provider Discovery Agent
Searches, filters, and ranks providers based on the patient's needs, preferences, location, insurance, and availability. Produces a scored shortlist using the Matching Engine.
Safety Agent
Runs alongside every workflow. Assesses severity tier (crisis, high-risk, elevated, standard), enforces scope-of-practice rules, and triggers escalation protocols when needed.
Appointment Booking Agent
Handles provider selection confirmation, slot availability, booking, rescheduling, and cancellation, with real-time calendar integration and patient preference matching.
Discovery to Shortlist Workflow
The patient journey begins with intake. The system captures the patient's needs conversationally (symptoms, preferences, past therapy experience, insurance, and scheduling constraints) and then routes this signal to the Provider Discovery Agent.
Patient Intake
The patient describes their needs through a conversational, supportive interface. The system captures structured signals from unstructured input, not a form, but an adaptive dialogue.
Signal Extraction
The patient's input is parsed into a structured Client Signal, including presenting concerns, severity indicators, therapeutic preferences, logistical constraints, and safety flags.
Provider Discovery
The Provider Discovery Agent queries the provider database, applies eligibility filters (insurance, location, specialization), and passes candidates to the Matching Engine.
Shortlist Generation
The Matching Orchestrator produces a ranked shortlist with explainable scores. Each recommendation includes why the provider was selected and how the fit was determined.
Provider Selection to Booking Workflow
Once the patient reviews the shortlist, the system transitions to the booking workflow, handled by the Appointment Booking Agent.
Provider Selection
The patient selects a provider from the shortlist. The system confirms the selection and checks real-time availability.
Slot Confirmation
Available appointment slots are presented based on patient and provider schedules. The patient confirms their preferred time.
Booking Confirmation
The appointment is booked and confirmed. Both patient and provider receive confirmation with session details, video link (via VideoSDK), and preparation guidance.
Payment to Invoice and Reminders Workflow
Payment and post-booking communication are handled as part of the agentic workflow, not bolted on after the fact.
Invoice Generation
On booking confirmation, an invoice is automatically generated based on session type, provider rate, and insurance applicability. Processed through a third-party payment gateway.
Payment Processing
The patient completes payment within the flow. The system tracks payment status and handles retries, failures, and receipts.
Reminders and Follow-Ups
Pre-session reminders, post-session follow-ups, and rescheduling prompts are sent via Firebase Cloud Messaging. Automated, timely, and contextual.
Matching Logic and Explainability
Provider matching is not keyword search. It is a structured, multi-dimensional reasoning process designed to produce the best therapeutic fit for each patient.
Matching Components
- Provider Genome: A structured profile of each provider: specializations, modalities, experience, availability, populations served, and session formats
- Client Signal: The patient's extracted needs: presenting concerns, severity, preferences, constraints, and prior therapy history
- Patient Profile Context: Longitudinal context from prior interactions, session history, and profile data
Matching Engine
- Therapeutic Fit Engine: Scores provider-patient compatibility across clinical, logistical, and preference dimensions
- Matching Orchestrator: Coordinates inputs from Provider Genome, Client Signal, and Profile Context to produce a ranked, explainable shortlist
- Explainability: Every recommendation includes a clear rationale: which factors contributed, how scores were weighted, and why alternatives were deprioritized
Safety and Clinical Governance
Safety is not a feature. It is an architectural requirement. The Safety Agent operates in parallel with every workflow, assessing risk and enforcing governance rules in real time.
Severity Tiers
Crisis
Immediate risk detected. The system halts the workflow, provides crisis resources, and escalates to a human clinical contact.
High Risk
Significant clinical indicators. The system restricts provider pool to appropriately credentialed specialists and flags for review.
Elevated
Moderate indicators present. The system adjusts matching weights to prioritize providers with relevant experience and monitoring capabilities.
Standard
No elevated risk indicators. The full provider pool is available, and matching proceeds with standard weights.
Governance Rules
- Scope-of-practice enforcement: Providers are only matched for services within their licensed scope
- Severity-tier rules: Higher-risk patients are routed to providers with appropriate credentials and experience
- Escalation protocols: The Safety Agent can interrupt any workflow and redirect to human oversight when clinical thresholds are met
Observability and Audit Trail
Every decision, state transition, and agent action is logged and traceable. This is not optional instrumentation. It is a core requirement for clinical and regulatory compliance.
Built for Audit
- Integrated with LangSmith for LLM call tracing, token usage, and response quality monitoring
- Full decision lineage, from patient input to provider recommendation to booking confirmation
- Designed to support clinical audit requirements and regulatory reporting
Technology Stack
The Psyter Patient Agent is built on a production-grade stack selected for reliability, observability, and clinical compliance.
Outcome
The Psyter Patient Agent delivers a care navigation experience that is qualitatively different from directory search or basic chatbot triage.
For Patients
- Conversational and supportive: Patients interact with a system that feels like a thoughtful guide, not a form or decision tree
- Safer by default: Every interaction is assessed for safety, and the system escalates before problems become crises
- End-to-end experience: From intake to booking to payment. One continuous, coherent flow
For the Platform
- Explainable and auditable: Every decision can be traced, reviewed, and reported, meeting clinical and regulatory requirements
- Modular architecture: Each agent can be updated, retrained, or replaced independently without disrupting the overall system
- Production-grade: Built for reliability, not just demos, with observability, error handling, and graceful degradation
Ready to Build Your Own Agentic System?
SaaStoAgent builds production-grade, multi-agent systems: controlled, auditable, and designed for your domain.