One platform. Everything personal stays on the device.
Enumora is built on a single architectural rule: what is personal never leaves the child’s phone; only what helps the system crosses — consented, minimal and signed. Everything else follows from that rule, including why it works offline.
enum enumora.platform {
- {e}.brain the on-device tutor
- {e}.edge the device surface
- {e}.pass verifiable credentials
- {e}.sense attention research
- {e}.labs vocational training
}
{e}.brain
A full tutor that fits in a family phone.
Small language models now run on a ₹10–15,000 Android phone. That shift is why Enumora can exist. {e}.brain combines a compact, quantised language model with Indic speech recognition and a warm neural voice — all running on the device, with no cloud call anywhere in the student app.
- A deterministic lesson engine. Scripts set the pedagogy; the model classifies answers, asks follow-ups and speaks naturally. The model never improvises the teaching.
- A misconception bank per topic. The errors children actually make, mapped chapter by chapter — the asset a chat window never accumulates.
- A latency budget. The tutor’s first word should follow the child’s last within about three seconds. A phone that can’t keep up moves to a smaller model and text-first mode, openly — never silently worse.
- Every release benchmarked on real reference phones, not emulators, before it reaches a child.
What runs on the phone
Language model (1–4 billion parameters, 4-bit quantised) · streaming Indic speech recognition · Indic text-to-speech · lesson engine · credential signer.
What it needs
Android 10 or later with at least 4 GB of RAM; 6 GB or more runs the larger model. Phones in the ₹10–15,000 range qualify. A shared family phone holds a separate, secret-protected profile for each child.
What it never needs
An internet connection during a lesson, a data pack, or an account with an email address.
{e}.edge
One app, two ways into a classroom.
The tutor, the privacy rules and the credentials are identical everywhere. What changes is whose device it runs on, and who pays.
Bring your own device
The Enumora app on the Android phones families already own. No hardware cost to anyone; a small set of loaner phones for children who need one. Syncs whenever the phone meets any permitted Wi-Fi.
{e}.edge tablets
Government tenders ask for device and platform from a single vendor. {e}.edge is a rugged, solar-friendly tablet with the full tutor built in, managed centrally and updated in staged rings — backed by outcomes data from the private-school pilots.
Photosynthesis — 84% mastery
- Learner
- Yuvaan N.
- Class · subject
- Class 7 · Science
- Chapter
- Ch 1 · Nutrition in plants
- Issued
- 17 Sep 2026 · Green Valley Public School
- Share code
- EPX-7K2M4Q-B7488V
Each credential is cryptographically signed. Years from now, an employer verifies any of them in under a second — without calling the school.
{e}.pass
Mastery that can be proved — by anyone, for life.
Marksheets can be lost, faked or misread. A credential from Enumora is signed by the phone at the moment the tutor verifies mastery, endorsed by the school on the next sync, and checkable by a stranger in under a second.
Signed on the device
The phone signs the credential with the learner’s own key the moment mastery is verified — even with no internet. The child earned it offline, so it exists offline.
Endorsed by the school
When the phone next syncs, the school adds its endorsement. Teachers can approve each one, or the school can endorse automatically.
Verified by anyone
A parent, a college or an employer enters the share code or scans the QR. The signatures are checked in under a second. If it isn’t signed, it doesn’t verify.
Built on open standards: W3C Verifiable Credentials 2.0 and Open Badges 3.0. DigiLocker issuance is on the roadmap.
Research tracks
Two layers we are researching in the open.
Neither ships in the pilot. We describe them because they shape the architecture today — and because we would rather be held to a published direction than surprise anyone later.
{e}.sense
In researchCan a tutor notice when a child is losing focus, and respond by slowing down, offering a hint earlier, or suggesting a break? Any such signal would be processed on the device, never transmitted, switched off by default, managed by parents, and shipped only after peer review.
{e}.labs
In researchIndia has 14,615 ITIs training the next generation of electricians, welders and fitters, largely on equipment that gives no feedback. {e}.labs retrofits cameras, sensors and edge compute onto standard workshop rigs so every trainee gets specific, real-time coaching — and a verifiable skills credential at the end.
Alongside what you already run
Keep your LMS. Enumora is a different layer.
A learning management system stores and plays content. Content doesn’t teach. Enumora is the tutor beside the child: it finds what a child has misunderstood, repairs it in the child’s language, schedules the revisit before forgetting sets in, and signs the credential when the concept is secured. Nothing you have invested in is replaced.
Engineering commitments
- No cloud inference in the student app, and no analytics SDKs.
- A whitelist, not a filter, decides what can leave a device: IDs, categories and bounded numbers only. Free text has no field to travel in.
- Mutual-TLS sync to infrastructure hosted in India.
- Standards-based integration with school systems is on the roadmap.
See the whole platform working — in a browser, offline.
Our product walkthrough shows every persona: student, teacher, parent, school and network. Ask us for an access code.