Skip to content

About

Public technical docs for TherapyFlow — AI clinical agent for psychologists. Stack, agentic layer, roadmap.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

5 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

TherapyFlow — Public Docs

AI clinical agent for psychologists. Sessions in. Compliant clinical notes out.

🌐 therapyflow.app

This repository hosts the public-facing technical documentation for TherapyFlow: architecture overview, agentic clinical layer spec, and roadmap. The product source lives in a private monorepo.


What TherapyFlow does

Psychologists spend 6–10 hours per week writing clinical notes after sessions. TherapyFlow eliminates that work by capturing the session natively and producing compliant clinical notes through an orchestrated agentic pipeline.

The therapist runs a desktop app during the session. After the session ends, the note is in their inbox before they finish their coffee. No typing, no templates, no transcription pasted into ChatGPT.


Stack

Layer Technology
Desktop client Tauri (Rust core + React/Vite UI)
Audio capture Native Rust → WAV writer + WebSocket streaming
Backend NestJS (TypeScript) on Heroku
Web app Next.js 15 (React 19) on Netlify
Voice worker FastAPI + PyTorch + SpeechBrain ECAPA-TDNN on Heroku
Database MongoDB Atlas
Audio storage AWS S3 (presigned PUT, never touches backend RAM)
LLM orchestration Claude (Anthropic) + GPT-4o (OpenAI), A/B routed
Speaker diarization Self-hosted SpeechBrain ECAPA-TDNN, 192-dim embeddings
Voice agents Vapi + ElevenLabs + TinyFish

Agentic Clinical Layer

Four agents run on top of the transcription pipeline:

  1. Note Generator — produces SOAP-format clinical notes from utterances.
  2. Clinical Progress Agent — tracks patient evolution across sessions (symptom trajectory, treatment adherence, risk signals).
  3. Stylistic Clone — learns each therapist's note-writing voice and applies it to generated drafts.
  4. Supervisor Mode — flags drafts that diverge from the therapist's evidence-based protocols.

Voice-first patient intake is handled by Sally, an AI receptionist: phone call → triage → therapist match → calendar booking → handoff.


Roadmap (public-facing milestones)

  • ✅ Sprint 1–2 — Recording architecture + WS streaming
  • ✅ Sprint 3 — Processing pipeline + speaker confirmation flow
  • 🚧 Sprint 4 — Founding Therapist Program + Paddle billing live
  • 🚧 Agentic clinical layer — Progress Agent + Stylistic Clone in production
  • 🔜 Global rollout — English-first product with localization for Spanish, Portuguese, and additional markets

Team

Jair Celis — CTO & Co-founder · @jaircelisv

2× founder. Founded Godat in 2016 — clinical claims and audit software still operating profitably, serving Allianz, Mapfre, and Sanitas. Now on Godat's board while building TherapyFlow full time.

13 years shipping production healthtech software, building from Colombia for the world. Built RIPS processing infrastructure (Colombian healthcare claims standard) at production scale — millions of records.


Contact

✉️ jair@therapyflow.app · 🌐 therapyflow.app

For investors, partnerships, and clinical pilots.

About

Public technical docs for TherapyFlow — AI clinical agent for psychologists. Stack, agentic layer, roadmap.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors