AI Therapy Platform for Substance-Use Disorder
An end-to-end AI platform for mental health and SUD therapy — agentic workflows, predictive relapse analytics, automatic session documentation, and patient + clinician apps.
The Challenge
A funded healthcare startup needed the full AI platform engineered from scratch: agentic workflows, predictive analytics for relapse detection, automatic session documentation, and integration with patient-mobile and clinical-web applications — all production-grade from day one.
The Architecture
What we built.
Agent Platform
LangChain + LangGraph multi-agent workflow — event-driven triggers, conversational + executor agents, retrieval + reasoning loops for clinician-patient orchestration.
Predictive Analytics
Classical ML + LLM hybrid models for early detection of relapse / clinical worsening in SUD populations, trained on structured + unstructured signals.
Session Documentation
Real-time Whisper + Whisper-Turbo transcription with speaker diarization (therapist vs. patient), automatic SOAP-style summaries, task extraction, follow-up drafting.
Business Value
What changed for the customer.
- MVP shipped in weeks rather than quarters — composable agent architecture made it possible.
- Therapist documentation load dramatically reduced via automatic session summaries.
- Predictive relapse detection in production pilot — proactive clinical intervention instead of reactive.
- Bilingual-ready architecture (English primary, Hebrew parity available) from day one.
