Models of how, not only of what.
Langevin molecular dynamics for a peptide on its target, delay equations for cortisol feedback, potential landscapes for mood. Machine learning where it earns its place, mechanism everywhere else.
Cognibrex Technologies · Computational neuroscience
Instruments for the living mind.
We build working, checkable models of the brain, from a six-residue peptide designed against amyloid-β42 to the language of a psychiatric note. Eight of them already run live in a browser. The hall is open to researchers by request.
- Live instruments
- 8all computing in your browser
- Scale covered
- 10⁻¹⁰ mto a clinical text corpus
- Sleep staging
- κ 0.798held-out subjects, one EEG channel
- Peptides scored
- 150AI-designed against amyloid-β42
- Uploads needed
- 0your recordings stay on your device
Overview
A computational lab for the brain, built to be checked.
Cognibrex Technologies is the research company behind Neulish. We turn questions in neuroscience and psychiatry into models you can run, perturb and audit, then build products from the ones that hold up.
Every number says where it came from.
Each instrument separates real inputs from modelled ones, shows how it was checked and states its limits, including the results that came out negative.
No server, no upload.
Filters, features, classifiers and simulations run on the visitor's own machine. A night of EEG is staged in seconds without leaving the laptop it was loaded on.
What we're building
The products in our pipeline.
Research products in development, with their status stated plainly. None of them is a medical device, and none makes a clinical claim.
CLARAAI-designed peptides against Alzheimer's disease
Generative design, structure prediction and multi-objective selection of short peptides that bind amyloid-β42, with their dynamics simulated live against the target.
NeuroScope SleepSingle-channel EEG sleep staging
A whole night staged from one EEG channel in seconds, at κ 0.798 on held-out subjects. It reads hospital polysomnography and consumer headbands, and flags the epochs a human should check.
Neulish AIPassive mood-episode and stress detection
Early warning of bipolar mood episodes and stress from speech emotion and wearables, with every detector first tested against simulated patients whose ground truth is known.
MindNERClinical language for psychiatry
Psychiatric entity extraction, negation, ICD-10-CM and DSM-5 coding, and de-identification for clinical notes and the research literature.
Brain dynamics modelsStress, mood and neurodegeneration
A hypothalamic–pituitary–adrenal model with cortisol feedback, a mood landscape that shows critical slowing down, and an amyloid–tau–neurodegeneration cascade projected with Monte-Carlo intervals.
NeulishThe product already in people's hands
A mental-wellness companion: AI voice and chat around the clock, clinical check-ins, and licensed psychologists when someone needs a person.
Request access
The hall opens by invitation.
The instruments are early and they are ours to explain, so we let people in one at a time. Tell us who you are and what you work on.
Request
Two minutes: your affiliation, your field, and what you would like to try.
Review
A person on the Cognibrex team reads every request. Nothing is automated.
Enter
Approved researchers get an Access ID and password by email, and the hall opens.
Request received.
Thank you. A person on our team will read it. If it's approved, your Access ID and password will arrive at .
Questions
What researchers ask first.
If yours isn't here, ask it in the feedback form below. The founding team reads every one.
What is Cognibrex?
Cognibrex Technologies is a computational neuroscience company in India. It builds working models of the brain at every scale, from the molecular dynamics of a peptide against amyloid-β42 to the language of a psychiatric note, and it is the company behind the Neulish mental-wellness app.
What is the Instrument Hall?
Eight live computational instruments that run in the browser, ordered by the physical scale they model, from ångströms to a clinical text corpus. They cover peptide molecular dynamics, neuron-level Alzheimer's pathology, the HPA stress axis, EEG sleep staging, neurodegeneration staging, bipolar mood dynamics, mood-episode detection and psychiatric language.
Who can get access?
Researchers, clinicians, students, research-led companies and investors. Fill in the access request with your affiliation and what you work on; a person on the team reviews it, and approved requests receive an Access ID and password by email.
Are these medical devices?
No. Every instrument is research software and every model is a computational hypothesis. None gives a diagnosis, and no peptide shown has been tested in a wet lab. Each instrument states which numbers come from real data and which from equations.
Does my data leave my device?
The instruments compute in your browser. A sleep recording you load into NeuroScope Sleep is filtered, featurised and staged on your own machine and is not uploaded.
How accurate is the EEG sleep staging?
On held-out subjects from the Sleep-EDF database, the single-channel model reaches a Cohen's κ of 0.798 against expert scoring. The browser version reproduces the Python reference exactly on a full test night (872 of 872 epochs).
Can we collaborate or validate a model with you?
Yes. Tick research collaboration or datasets and validation in the access request, or send a note through the feedback form. Wet-lab, clinical and dataset partners are especially welcome.
Feedback
Tell us what to build next.
An idea, a dataset we should test on, a model we got wrong, or a question you want a simulation for. No account needed.
- Read by the founding team, not a bot.
- Leave an email only if you want a reply.
- Critical feedback is the most useful kind.
Thank you.
It's with the founding team now. If you left an email, expect a reply from a person.