web embed
Real EEG at real speed: OpenNeuro ds002778 (CC0), harmonised to ESF
Review
Vendor EEG, in the browser
MIND
Intelligence that abstains
ESF
Open preprint and code
ENCEPHLIAN logo
by Aposematium · Hyderabad
ENCEPHLIAN
Scalable harmonization infrastructure for clinical EEG.
We turn recordings from the EEG machine a clinic already owns into one trusted signal, then add narrow intelligence that knows its limits.
The problem
EEG is locked in, read by hand, and models break.
What neurologists told us
“I spend hours on screen pressing the same key, page after page, to find a few spikes. It’s exhausting, and a hard case can take days, sometimes a week, to report.”
Consultant neurologist, large hospital, Hyderabad
“I have very little time to read EEGs, so I read them case by case. A lot of opportunity is missed.”
Epilepsy neurologist, 26 years in practice, Hyderabad
Locked in
Most EEG makers store recordings in their own closed format. Older, often offline machines get no new analytics.
Read by hand
A neurologist scrolls 20 to 40 minutes of signal to find the few seconds that matter, then types the report.
Few readers
Few neurologists are trained to read EEG, so their time is the bottleneck.
Models do not travel
A model trained on one recorder can fail, often silently, on another recorder's sampling rate, gain or mains.
Measured on slide 3 →
ENCEPHLIAN logo
ENCEPHLIAN
02
Our evidence · ESF preprint
Change the sampling rate and good EEG models fail.
Three frozen public EEG models, normal vs abnormal, on held-out TUH recordings (n = 276), then the same recordings resampled to 512 Hz (AUROC). ESF restores all three.
web embed
0.90
Three public EEG models on clean recordings: 0.89 to 0.90.
0.40
BIOT at 512 Hz, rate not corrected. Worse than a coin flip.
0.90
Harmonized with ESF. Gain and mains shifts are fully undone too.
ENCEPHLIAN logo
ENCEPHLIAN
03
Preprint 10.5281/zenodo.22772008 · 3-seed mean · limit: at 128 Hz only EEGPT recovers
How it works
Fix the signal first, then add narrow intelligence.
Our core question: what is the minimum representation of brain activity that clinicians and machines can both trust?
web embed
ENCEPHLIAN logo
ENCEPHLIAN
04
ESF v1.0: 256 Hz, µV. The Zenodo preprint tests its 250 Hz, robust z-score form
Product · Review
Vendor EEG in the browser. Nothing uploads.
Natus .e, Nihon Kohden, BDF, EDF, EDF+ and EDF+D open on the clinic's own computer. No vendor software, nothing installed.
Screen recording of ENCEPHLIAN Review opening a public EEG file, checking it and showing the signal-quality map
Real 10-second screen capture of a public recording (OpenNeuro ds002778, CC0). Research preview, not for clinical use.
Checks the file first
Channels, sampling rate and mains, checked before anything else.
Signal-quality map
Which electrodes and minutes are readable, at a glance.
Whole recording, then the report
Jump to the minutes that matter, then write and sign.
ENCEPHLIAN logo
ENCEPHLIAN
05
Product
MIND
Narrow intelligence that knows when to stay silent.
Points the neurologist to possible seizures and discharges and suggests an impression. The neurologist accepts, edits or rejects.
Screen recording of MIND listing seven things to look at, a possible seizure accepted, and a suggested impression
Real 10-second screen capture: MIND on a held-out public TUH recording. Research build, not for clinical use.
web embed
How to read these: AUROC: 1.0 always tells the two apart, 0.5 is a coin flip. F1 is stricter: 1.0 finds every seizure with no false alarms, and the best open model scores 0.44.
Abstains by design. It reports only on recorders it has been measured on, so far at native 250 to 400 Hz. Otherwise, or when unsure, it stays silent.
ENCEPHLIAN logo
ENCEPHLIAN
06
Internal held-out TUH test, run once · not externally validated · verify on request: 10.5281/zenodo.23194286
Product
MIND
Machine Intelligence for Neural Data
web embed
Held-out TUH patients
seizure event F1 · 55 patients · opened once
Another hospital: Siena
seizure event F1 · 14 patients · 512 Hz · public data
Discharges and abnormal recordings: held-out results on slide 6; on validation patients MIND is level with the best open model on both. Deployed MIND stays silent on 512 Hz recordings until measured more widely.
ENCEPHLIAN logo
ENCEPHLIAN
07
Comparators scored after the opening, from the same frozen predictions · verify on request: 10.5281/zenodo.23194286
On held-out patients, MIND finds the most seizures.
Patients kept out of every design choice, against the four open models MIND is built on.
MIND
Open EEG foundation models MIND is built on, each with our head: LaBraM, CodeBrain, AXON-EEG, MANAS-2
The opening · CDSCO, India's device regulator
0
EEG event-detection software products licensed in India.
EEG machines are licensed. Software that marks seizures and spikes is not.
Machines, not software
RMS, Medicaid Systems and S.S. Medical hold Class B licenses for EEG machines.
Not on the list
Persyst, encevis and BESA, used in hospitals abroad, are not licensed here.
We have filed
Test-license applications: Review as Class B, MIND as Class C. October 2026.
ENCEPHLIAN logo
ENCEPHLIAN
08
Our check of CDSCO's public list of approved devices, 6 October 2026 · anyone can repeat it
Distribution and data
Neurologists who sign, and EEG distributors.
Neurologists
Letters of intent to use and pay. One is our partner clinic, testing on site.
2 signed
Nihon Kohden distributor
Supplies Hyderabad clinics and hospitals.
Agreed to help
Indian EEG makers
RMS, Clarity Medical and Allengers.
In talks
Why Indian data: MIND learned from one US hospital. Next: two neurologists read 100 studies per recorder, and MIND is measured on each.
Distributors
Clinics
Signed reports
Consented Indian EEG
Models built for India
ENCEPHLIAN logo
ENCEPHLIAN
09
The Indian corpus is a plan: opt-in, consented, targeted to begin in Q4 2027
Defensibility
Hard to copy, and open for anyone to check.
Six file readers
Natus .e, Nihon Kohden, BDF, EDF, EDF+ and EDF+D, decoded in the browser. Each closed format is slow, careful work.
Open and reproducible
ESF preprint and code are public on Zenodo (FAIR). Anyone with TUH access can rerun the public‑data results.
Measured domain that grows
MIND speaks only where it has been measured. Each new site and recorder adds evidence a rival must repeat.
Regulatory head start
Test-license applications filed for Class B and Class C, in a category with no licensed EEG event‑detection software.
MIND review list in the product
MIND's review list: each item carries its confidence, and unvalidated items say so.
ENCEPHLIAN logo
ENCEPHLIAN
10
MIND verification record · verify on request: 10.5281/zenodo.23194286
Landscape
Others make recorders, ICU tools or models. We connect the recorders to the models.
Who
Examples
What they do
For us
Hospital EEG software, abroad
Persyst, encevis, BESA
Spike and seizure detection on hospital systems
Not licensed in India
ICU seizure monitoring, US
Ceribell, Piramidal
Monitoring in intensive care
A different setting
EEG foundation models
NeuroDx (now Mannas AI): MANAS‑2, AXON-EEG
Open pretrained models
MIND builds on them
Indian EEG makers
RMS, Medicaid Systems, S.S. Medical, Clarity Medical, Allengers
EEG machines
Possible channels and partners
Us
Review · MIND · ESF
Harmonization in the clinic's browser; narrow intelligence on a server in India
Connects recorders to models
ENCEPHLIAN logo
ENCEPHLIAN
11
License status from our check of CDSCO's public list, 6 October 2026
Market · our clinic census
One city counted by hand, scaled to India.
100
EEG sites in Hyderabad, from one-neurologist clinics to hospital labs
₹4.5 cr
a year in Hyderabad if every EEG is reported through us at ₹150, about 10 a day per site
₹50–90 cr
a year across India's ~1,100 to 2,000 EEG sites with Review alone; ₹100 to 175 crore with MIND
What they record on
Most clinics
Clarity and RMS portables
Review opens their EDF export today.
Hospital labs
Nihon Kohden
Mid-size and large hospitals. Read natively.
A few sites
Natus and Nicolet
Read natively.
Review opens files from every recorder brand in the census, today.
ENCEPHLIAN logo
ENCEPHLIAN
12
India: census scaled by neurologist counts (IAN ≈3,000; listings ≈4,700, 242 in Hyderabad) · 300 days
Business model · two licenses, two revenue layers
Class B pays first. Class C compounds.
Class B · Review · license Q2 2027
₹150
Per signed report
About ₹4.5 lakh a year from a typical site in our census.
Near-zero cost to serve: it runs in the clinic's browser.
A known path: EEG analyzers are already licensed in Class B.
Class C · MIND · target Q1 2028
₹3,000
A month, 10 reports
Overnight or immediate reads, priced by compute.
Tiers to be set with our first sites.
No EEG event-detection software is licensed in India. The first approval becomes the reference every rival must match.
Today: Evaluation only, with no sales before the Class B license. About 10 sites take us to cash-flow positive (our estimate).
How we scale: Bottom-up. Through the recorders clinics already own and their distributors. One Class C license opens every neurology clinic in India.
ENCEPHLIAN logo
ENCEPHLIAN
13
Working prices: Review ₹150 ±10% a report; MIND from ₹3,000 a month · 10 EEGs a day, 300 days
Status and roadmap
Class B and Class C start now, in parallel.
This round · 18 to 24 months of runway
Q4 2026
Q1 2027
Q2 2027
Q3 2027
Q4 2027
Q1 2028
Q2 2028
Evaluate
Test licenses, then demos
Shown to clinicians and evaluators, no claims
Adjudicated evaluation
2 readers · 100 per recorder
Paper: MIND on Indian EEG
Cross-vendor
Class B
Class B dossier and QMS
ISO 13485 · IEC 62304
Class B license, then sales
Review · ₹150 per signed report
Class C
Class C dossier
With an experienced Class C filer
Clinical investigation
Filed Q1 · reading study
Class C license, then sales
CDSCO approvals
MIND GTM
Funding
BIRAC BIG and grants
Jan 2027 call
Today
First revenue
Class C license
ENCEPHLIAN logo
ENCEPHLIAN
14
Aposematium Private Limited · DPIIT-recognized · TRL 5 · no revenue yet · dates after Today are targets
The ask
$250,000
turns internal results into external evidence.
web embed
Milestones
Class B license · Q2 2027
First revenue · Q3 2027
Class C license · Q1 2028
ENCEPHLIAN logo
ENCEPHLIAN
15
Convertible into Aposematium Private Limited · burn about ₹6 lakh ($7k) a month · 18 to 24 months of runway
Team
Hitesh Dammu
Hitesh Dammu
Co-founder, Product and Engineering
MS Biomedical Engineering, Arizona State University, 2023. Wrote the ESF preprint. Built ESF, Review, MIND and the regulatory files.
Ram Waddepally
Ram Waddepally
Co-founder, Operations, Sales and Marketing
BBA, IBS Hyderabad, 2022. Leads clinic outreach, the clinic census, evaluation sites and clinician relationships.
Clinical advisors · names on request
Epilepsy Neurologist
26 years in epilepsy practice. National hospital chain, and owner of our partner clinic, a high-volume neurology clinic.
Signed LOI · testing on site
Consultant Neurologist
Large multi-specialty hospital, and his own practice in Hyderabad. DM (Neurology), SCE (UK).
Signed LOI
Senior Neurosurgeon
Large multi-specialty hospital, Hyderabad. MCh; 5,000+ surgeries; neuromonitoring and medtech design.
Advisor
ENCEPHLIAN logo
ENCEPHLIAN
16
Aposematium Private Limited, Hyderabad · ESF, Review and MIND built in-house, on open foundation models
web embed
ENCEPHLIAN logo
ENCEPHLIAN
Clinical EEG Workload Cloud
Any recorder in.  ·  One trusted signal out.  ·  Intelligence that knows when to stay silent.
Check our work: preprint 10.5281/zenodo.22772008 · code 10.5281/zenodo.22771925 · MIND, on request: 10.5281/zenodo.23194286
Hitesh Dammu · +91 95157 18969
Ram Waddepally · +91 79931 13166