Zyora Labs
AI Models

Models we train & fine-tune.

From purpose-built education tutors to foundation models trained from scratch and applied research models — Zyora Labs documents model scope, evidence and limitations openly.

Healthcare · ECG research2026
Research model · Open source

VitalSync ECG — A context-aware model for annotated ECG beat classification

VitalSync ECG classifies annotated single-lead ECG beats into N, S, V and F groups using the current beat, two preceding beat crops and causal RR-history features. It is an openly documented research artifact, not a clinical monitoring or alert system.

ECGBeat classificationContext-aware ResNetMIT-BIHINCARTTorchScript
View model cardGitHub
0.4453
Validation macro-F1
4
AAMI-style beat groups
5.3 MB
TorchScript model
A100
Verified training hardware
Education · Fine-tuned2026
v1.0 · ZyoraByte

Byte-12B — A fine-tuned AI model for engineering education

Byte-12B is ZyoraByte's own AI tutor — a 12-billion-parameter model fine-tuned from Gemma 4 12B to teach across all 24 engineering streams. It explains concepts patiently, works through real university exam questions, and adapts to how each student learns.

EducationAI tutorGemma 4QLoRAExam prepEngineering
View model cardWebsite
12B
Parameters · Gemma 4 base
24
Engineering streams covered
4,096
Token context window
QLoRA
4-bit fine-tune on H100
Foundation · Pre-trained2026
Foundation model

AQ-5B — A foundation model by Zyora Labs

AQ-5B is Zyora Labs' own foundation model — a decoder-only Transformer trained from scratch in pure PyTorch to understand every educational stream, from Physics and Maths to Botany, Tamil and Engineering. It is the base that will power the next generation of ZyoraByte and Zyora AI.

Foundation modelPre-trainingPure PyTorchRoPE · RMSNorm · SwiGLU · GQAMultilingualEducation
View model card
5B
Parameters
32K
BPE tokenizer vocabulary
EN · TA
Bilingual — English & Tamil
Scratch
Pre-trained from zero
How we build models

Our model principles

01

Built for every stream

Our education models are department-agnostic by design — from mechanical and civil to ECE and CS. No stream is a second-class citizen.

02

Trained on real material

We fine-tune on genuine university coursework and past exam papers, so answers match what students are actually assessed on.

03

Owned end to end

From datasets to fine-tuning to serving, we control the full pipeline — including our own foundation models trained from scratch.

Want early access or to collaborate?

We work with institutions, researchers and partners building the next generation of education AI.

Get in touch