Company
AI that experts can work with.
NeuroLink AI is an international team. We build cognitive AI for high-stakes work in finance and healthcare: systems that experts can follow, question and steer.
Why we exist
The gap we set out to close.
Models became accurate faster than they became understandable. In high-stakes work, accurate is not enough.
We started from a simple observation. A system can be right and still be of little use to the person who must answer for the decision. That person has to follow the reasoning, question it and sign it.
When the reasoning is hidden, experts check everything again, and the time the system saved is spent on verification. When the system cannot see what the expert means, every intention has to be translated into a prompt.
NeuroLink AI exists to close that gap from both sides: systems that show their reasoning in the terms experts already use, and that take in intent, rules and judgement as structure, not only as text.
The name
One name, two directions.
NeuroLink stands for the link between neural networks and human reasoning. The link has to carry meaning both ways.
AI to people
Neural networks
Learn from data
People to AI
Human reasoning
Judge and decide
Neural networks learn patterns from data. People reason about causes, rules and consequences. We build the link between the two.
- AI to people
Reasoning you can follow
Every recommendation arrives with its evidence, the rules it applied and the steps it took, in the terms the expert already uses.
- People to AI
Intent the system understands
Goals, constraints and expert judgement reach the system as structure it can represent, follow and learn from. Not only as a prompt.
- In between
A shared reasoning layer
Learned patterns, explicit rules and a model of how the expert reasons meet here, so that people and system work from the same picture of the decision.
Mission and vision
Mission and vision
- Mission
- Make AI a reasoning partner for experts in high-stakes work.
- Vision
- Every important AI-assisted decision can be followed, questioned and improved by the people responsible for it.
Principles
Five principles we build by.
They apply to our research, to our software and to how we work with partners.
Adapt the machine, not the human.
Systems should fit how experts think.
Reasons before answers.
A recommendation without its reasoning is a liability.
People stay in command.
AI proposes. Accountable experts decide.
Earn trust with evidence.
Show the sources, rules and confidence behind every recommendation.
Research-grade, production-hard.
Ideas from cognitive science, held to the standards of systems that run every day.
How we work
Research and engineering, together.
One team for research and engineering
The people who study how experts reason also build the systems. An idea is ready when it works in a real workflow, under real constraints.
With partners
We develop CogniFlow with teams in finance and healthcare, on decisions they make today. Their experts tell us where the system helps and where it fails them.
Inside their workflows
The system goes where the work already happens and takes the shape of how the experts there reason. They should not have to change how they think to use it.
Team
An international team.
We work across countries and time zones, and meet partners online or in person where it helps.
Careers
Open roles.
No open roles listed right now. If you work on cognitive architectures, neuro-symbolic machine learning or human-AI interaction, write to us.
Questions
Questions about how we think.
Short answers about our concepts and how we work.
What is cognitive AI?
Cognitive AI is AI built around how people think. It models what an expert pays attention to, which rules they apply and how they weigh evidence, and it works in that shape. The aim is a system whose reasoning a person can follow and whose behaviour a person can steer.
What does NeuroLink mean?
It stands for the link between neural networks and human reasoning. The link runs both ways. From AI to people it carries reasoning you can follow. From people to AI it carries intent the system understands. A shared reasoning layer sits between them. More about the name.
What is neuro-symbolic AI, and why combine the two?
Neural networks learn patterns from data and read language well, but their reasoning is hard to inspect. Symbolic methods hold explicit rules and constraints that people can read and check, but they do not learn from raw data. Combined, the neural part finds what matters in documents and history, and the symbolic part applies the rules and policies a decision must respect, in a form people can inspect.
How can an AI system understand human intent?
Our approach does not rely on a prompt alone. Goals, constraints, priorities and expert judgement are captured as explicit structure: the goal behind a request, the rule that matters in this case, a preference for one kind of evidence. The system can represent that structure, act on it and show where it did. When an expert corrects the system, the correction is recorded with its reason and used for the next case.
Does the AI make the decision?
No. People decide. The system prepares a recommendation, shows its evidence and the rules it applied, and waits for a person to approve or override it. Nothing is final until someone accountable signs it. An override is recorded with its reason, so that the system can learn from it.
What is CogniFlow, and can we use it?
CogniFlow is the decision-support layer we are building. It sits inside existing expert workflows: it prepares a decision, shows its reasoning, takes in what the expert intends and learns from the people who sign it off. It is in development with partners. If you would like to work with it early, join the partner programme. More on the CogniFlow page.
Which industries do you work in?
We start in finance and healthcare, where decisions are frequent, consequential and reviewed. The same approach fits any work where decisions must be explained to a reviewer, an auditor or a regulator, such as insurance claims or public administration. In healthcare, our systems support documentation, coordination and planning. They do not diagnose or treat. See solutions.
How do you handle sensitive data?
CogniFlow is designed to run where your data already is, in your own cloud or on-premise. Before any data is shared with us, we agree in writing what we may use, where it is processed and who can see it. We do not need data to start a conversation.
How can we work together?
Start with a conversation. If your experts make a decision often and you would like to shape CogniFlow, apply to the partner programme. For research collaboration, press or anything else, write to us or email [email protected].
Talk to us about a decision.
We work with a small number of partners in regulated, high-stakes work.