Assess
Before a session begins, Neuro checks whether the user is appropriate for an AI-led session. If not, it refuses and routes the user toward human support.
Working software. Working hardware.
Closed-loop AI
for mental health.
A voice AI that conducts structured mental health sessions and uses physiological feedback to understand how the person is actually responding.
2,113daily users
98%voice sessions
$0marketing spend
Current users are in Russia and use the product in Russian. Payments have not been enabled yet.
Session active
12:48
“Take your time. What changed when you said that?”
Physiological arousalElevated
Passive sensing prototypeFrontal electrical signal + EDA
Sixty seconds
The product in one minute.
The founding belief
Mental health still runs on self-report.
We are building the missing feedback loop.
A person can say “I’m fine.” Today, a human professional or AI has to infer what is actually happening. Neuro adds synchronized physiological feedback to the conversation, so the system can observe a reaction, adapt, and measure what happens next.
How it works
One continuous loop, not a dashboard.
Talk
A voice AI conducts a structured session. Hundreds of agents cross-check safety, context, and compliance with a clinician-designed protocol.
Measure
The sensing band records physiological arousal and forehead tension while the session is happening.
Adapt
The system can slow down, pause, stay with a topic, or change technique, then measure the response to that decision.
Voice+Physiology→State estimate→AI response→Measured reaction
Already live
Demand appeared before monetization.
Current product usage in Russia.
A behavioral signal, not a claimed clinical outcome.
Voice won decisively over text.
All current growth has been organic.
50+
Unexpected user segment
Voice unlocked an audience we did not target.
People over 50 became Neuro’s largest segment. Speaking is easier and less embarrassing than typing for users who are uncomfortable with apps or written communication.
The sensing layer
Built for conversation, not meditation.
Speaking contaminates frontal EEG with facial and jaw-muscle artifacts. Neuro therefore combines frontal electrical sensing with skin conductance, which remains usable while a person is speaking.
- Signal stackFrontal electrical sensing + EDA
- ElectrodesDry, user-wearable
- ConnectivityBluetooth
- Weight70–80 g
- StimulationNone
Passive sensing only.No tDCS, tACS, CES, or other electrical stimulation circuitry.
Prototype statusBuilt and working
Safety by architecture
An AI that knows when not to talk.
General-purpose chatbots are designed to continue. Neuro assesses the user before a session and continuously checks whether the session should proceed. When the situation is outside its scope, it stops and directs the user to human support.
In one real case, the system refused to proceed when it detected symptoms consistent with an active panic attack and directed the user toward professional help.
What is different
The signal is not an output.
It is an input that changes the conversation.
- Talk to the user
- Rely mostly on self-report
- Do not use synchronized physiology
- Conducts structured voice sessions
- Reads synchronized physiological response
- Adapts the session and measures what follows
- Measures signals
- Produces scores and feedback
- Does not conduct a live structured session
The compounding asset
A paired dataset of decisions and reactions.
Each sensor-enabled session can create a sequence that no standalone chatbot or wearable collects.
- 01What the user said
- 02What the AI did
- 03How physiology reacted
- 04How the session adapted
- 05What happened afterward
The conversational side exists today. The synchronized physiological side begins accumulating when devices reach users.
Business model
One membership.
Software and hardware together.
3 voice sessions per month
No device
Unlimited sessions
Sensing device included
What we test next
We have not yet proven that physiology improves outcomes.
Our first controlled study will compare identical sessions with and without the real signal reaching the model. If the signal does not improve outcomes or retention, the core voice product still works without it. We intend to learn this within months, not build for years around an untested claim.
Team
Consumer, engineering, clinical, and creative expertise.
AK
Andrey Kuzmin
Co-founder & CEOBuilt a consumer fashion brand to $3.6M in annual revenue. Previously founded a vertical SaaS company and an engineering firm with 100+ employees.
DK
Dmitriy Kareisha
Co-founder & CTOThree patents, 35+ shipped hardware projects, revenue-generating AI products, and builder of Neuro’s multi-agent architecture and sensing device.
OS
Olga Sagalaeva
Clinical LeadPracticing psychologist with 20+ years of experience. Designed the structured protocol, measurement framework, and safety boundaries.
YL
Yulia Litvinova
Product & CreativeBased in San Francisco with a US green card. Leads product design and produces AI-native content and advertising.
Neuro