In Session Lab
Google Cloud & Vertex AI Architecture

Platform Technology & AI Specifications

Fulfilling Google Cloud’s Omni-App vision—executing a continuous, full-duplex Perceive → Reason → Express loop for clinical psychotherapy simulation.

Enter Clinical Simulator View llms.txt

Product Description

In Session Lab (https://getinsession.com) is an AI-native clinical flight simulator engineered for psychotherapists, graduate trainees (MSW, Counseling, and PsyD/PhD psychology students), and academic clinical training institutions.

While conventional clinical education relies on high-friction peer role-plays or expensive standardized actors, In Session Lab provides a zero-stakes, interactive sandbox where clinicians practice high-stakes interventions before facing real clients in crisis.

Core Capabilities

Unscripted, Multi-Turn Simulation

Conversational, multilingual virtual clients exhibiting authentic clinical pushback, avoidance, and emotional dysregulation across core evidence-based modalities (ACT, DBT, CBT, SFBT).

Longitudinal Memory Layer

Simulated clients retain treatment history, dialogue breakthroughs, and assigned homework across multi-session arcs.

Objective Pedagogical Analytics

Real-time generation of model-fidelity scorecards, structured session notes, and standardized supervisory rubrics (SW-OSCE, C-OSCE, P-OSCE, and CCS-R).

Zero PHI & Zero Privacy Risk

Because all client personas and clinical scenarios are 100% synthetic, there is zero Protected Health Information (PHI) or privacy liability. This enables unrestricted session recording, rubric analytics, and repeatable high-stakes practice without regulatory friction.

By pairing clinical fidelity with automated evaluative scoring, In Session Lab bridges the gap between academic theory and real-world clinical competence—built by a practicing therapist for therapists.

Technical Specifications & Google AI Integration

In Session Lab fulfills Google Cloud’s Omni-App vision, built entirely in Google Antigravity (following proof-of-concept in Gemini Opal and prototyping in Google AI Studio), it executes a continuous, full-duplex Perceive → Reason → Express loop for clinical psychotherapy training.

1. Full-Duplex Audio Streaming (Gemini Live & Vertex AI)

We leverage the Gemini Live API via bidirectional WebSockets (BidiGenerateContent) for sub-second voice interaction. Sample-accurate Web Audio timeline scheduling and intelligent barge-in detection enable therapists and virtual clients to converse, pause, and interrupt organically in real time.

2. Webcam Telemetry & Somatic Presence

Periodic webcam snapshots evaluate the therapist’s eye contact, posture, focus, and facial expression. This visual telemetry is synthesized into a Somatic Presence score, integrating real-time non-verbal awareness into the multimodal loop.

3. Cognitive State Engine & Dynamic Resistance

Somatic Presence is weighted alongside vocal empathy, conversational pacing, and clinical model adherence (ACT, CBT, DBT). This engine incrementally updates the virtual client’s Energy Dial and Composite Trust Factor, dynamically injecting system directives to simulate authentic psychological defense or vulnerability.

4. Two-Tier Memory with Google Cloud Firestore

Ephemeral In-Session State: Manages live prompt anchors, sliding turn context, and relational trajectory.

Longitudinal Rehydration: Firestore persists multi-session therapeutic rapport, transcripts, and competency rubrics, rehydrating longitudinal patient memory across multi-week training arcs.

5. Autonomous Multi-Agent Faculty Debrief

Post-session, an autonomous multi-voice supervisor panel evaluates clinical performance, generating modality analytics and conducting interactive spoken Q&A debriefs.

In Session Lab exemplifies how Google’s end-to-end AI ecosystem powers production-grade, multimodal agentic training.