Early cognitive screening through natural conversation
Nöra is a real-time voice AI agent that conducts standardized cognitive tests through natural dialogue — making Alzheimer's pre-screening accessible to anyone, anywhere, in any language.
What is Nöra?
Nöra is an AI-powered cognitive screening assistant that talks with users naturally — just like a clinician would — and guides them through 4 standardized tests to assess cognitive health.
Users simply speak to Nöra. The agent listens, asks follow-up questions, shows images, and records responses. All scores are computed deterministically in the backend — the AI never calculates, it only collects.
4Cognitive Tests
8+AI Agents
MultiLanguage
Who is this for?
👴
Elderly individualsAt-home screening without a clinic visit
👨👩👧
Family membersEarly warning for loved ones' cognitive health
🏥
Healthcare providersPre-screening triage before clinical assessment
🌍
Underserved communitiesMulti-language access regardless of geography
Why It Matters
Social impact & public health value
Early detection of cognitive decline is a global public health priority.
🌍
55M+ People with Dementia
75% remain undiagnosed. Nöra makes pre-screening accessible from home.
⏳
Early Detection Saves Years
Early intervention can delay Alzheimer's progression by 3–5 years, improving quality of life.
💰
$1.3T Global Cost by 2030
Dementia care costs are staggering. Accessible screening reduces late-stage burden.
🏥
Reduces Clinical Burden
Pre-screening at home frees clinician time for patients who truly need in-person evaluation.
🗣️
Eliminates Language Barriers
Multi-language voice interaction means cognitive screening is not limited by geography or literacy.
🤝
Empowers Families
Families get objective, early data to discuss with doctors — removing stigma from seeking help.
How It Works
Simple as having a conversation
The user talks to Nöra naturally. Nöra guides through 4 tests and delivers scored results.
1
🔐
Login
Sign in and start a new screening
2
🎤
Talk to Nöra
Real-time voice conversation, interruptible
3
🧪
4 Tests
Verbal, memory, visual, orientation
4
📄
Results
Risk assessment + PDF report
Cognitive Tests
4 standardized screening tests
Each targets a different cognitive domain. All scored by backend algorithms — never by the LLM.
🗣️
Verbal Fluency
Say as many words as possible starting with a given letter in 60 seconds.
60s timed
📖
Story Recall
Nöra tells a story, user retells it. Backend performs similarity analysis.
Gemini 3.1 Flash Lite
🖼️
Visual Recognition
Nöra shows 3 dynamically generated images. User identifies each.
Imagen 4
🧭
Orientation
Nöra asks 7 time/place questions: date, day, city, etc.
7 questions
Technology
Built with Google AI & Cloud
Production-ready, fully deployed on Google Cloud Platform.
🤖 Gemini Live API
🎨 Imagen 4
⚡ Gemini 2.5 Flash
📝 Gemini 3.1 Flash Lite
⚛️ React + Vite
🚀 Node.js + Express
🗄️ PostgreSQL + Prisma
🐳 Docker
☁️ Cloud Run
🔐 Secret Manager
📦 Artifact Registry
🗃️ Cloud SQL
Architecture
System overview
Multi-agent orchestration with deterministic scoring and real-time WebSocket streaming.
User → React Frontend (Cloud Run) → Node.js Backend (Cloud Run) → Gemini Live API · Imagen 4 · Cloud SQL · Secret Manager
Key Design Decisions
Why it works
Critical architectural choices for reliability and clinical safety.
🎯 Deterministic Scoring
The LLM never calculates scores. All scoring is algorithmic in backend — zero hallucination risk in medical assessments.
🔌 WebSocket Proxy
Node.js proxy bridges browser ↔ Gemini Live API, handling auth, CORS, and session persistence server-side.
⏱️ Server-Side Timers
LLMs can't track time reliably. Test timers run on backend, events injected into the live session.