The intelligence behind
the handover.

CascadeAI orchestrates six specialized AI agents powered by Meta Llama, combining real-time speech processing, strict clinical protocol matching, and Care System data synchronization with typical response times under two minutes.

System Architecture

A unified data flow from clinical frontend to secure multi-agent AI orchestration.

🖥️

Frontend Client

React SPA + WebRTC Audio

Vercel
⚡

FastAPI Gateway

Stateless Python Orchestrator

UK Cloud Hosted
Deepgram Nova-2 Medical

Real-time WebM to Text
GDPR-compliant transcription

🧠 6-Agent Engine
Intake Update Verify Protocol Draft Coord
Meta Llama 3.3 70B
🗄️ PostgreSQL Care Records

Resident master records
Row-Level Security

Supabase

The Stack

Built on modern, scalable primitives for maximum reliability.

⚡

Frontend

React + MediaRecorder API

Real-time Base64 audio encoding in the browser.
⚙️

Backend

Python FastAPI

High-performance ASGI server orchestrating 6 agents.
🧠

AI & ML

Meta Llama & Deepgram

Meta Llama 3.3 70B reasoning combined with GDPR-compliant clinical transcription.
🗄️

Database

Supabase (PostgreSQL)

Structured Care Record data with strict Row-Level Security.
☁️

Deployment

UK Cloud Hosted

Vercel (UI) + Cloud API (Backend).

Multi-Agent Workflow

A deterministic state machine orchestrating non-deterministic models.

Audio / Text Input
→
IntakeAgent
Transcription & Extraction
→
UpdateAgent
Classification & Formatting
→
CoordinatorAgent
Orchestrator · Risk Scoring
→
VerificationAgent
Care Record Verification
ProtocolAgent
Clinical Compliance
→
DraftGenerator
Narrative & Timeline

Real-Time Data Pipeline

From spoken word to structured Care System insertion in under two minutes.

🎤
Care Team Audio
📡
Base64 Encode
🎵
ffmpeg (WAV)
📝
Deepgram Nova-2
🧠
Meta Llama JSON
🗄️
Row-Level DB

Database Substrate

PostgreSQL schema engineered for Care Record data integrity and clinical workflow tracking.

📇 public.residents
id uuid PK
resident_id text UNIQUE
name text
age integer
medications jsonb
vitals_history jsonb
🏥 public.staff_shifts
id uuid PK
staff_id text
shift_type text
resident_ids text[]
status text
📋 public.resident_updates
id uuid PK
shift_id uuid FK
resident_id text FK
update_type text
transcription text
extracted_data jsonb
record_verified boolean
📄 public.draft_handoffs
id uuid PK
shift_id uuid FK
status text
structured_data jsonb
narrative_summary text
📨 public.sent_handovers
id uuid PK
draft_id uuid FK
from_staff_id text
to_staff_id text
status text