arXiv preprint · 2025
When Validation Fails: Cross-Institutional Blood Pressure Prediction and the Limits of Electronic Health Record-Based Models
Azam, M.B. & Singh, S.I. ↗
131K+
ECG-Linked Patient Records
MIMIC-IV research cohort
4
Research Datasets
MIMIC-IV · eICU · AI-READI · D1NAMO
48
Model Configurations
SAGE-Net evaluation
2
Publications
arXiv · BMC (2025–2026)
What we're building
An independent healthcare AI startup.
Routine glucose monitoring usually means a finger-prick or a sensor inserted under the skin. Daibeats is investigating a different route: estimating glucose from the electrocardiogram (ECG) — a signal already recorded by hospital monitors, clinical ECG machines, and a growing number of wearables.
We are building software, not a new sensor. Our goal is a glucose-estimation tool that works from existing ECG data, starting with research and clinical partners and moving towards a practical healthcare product as the evidence matures.
Daibeats technology is under research and validation. It is not a medical device and has not received regulatory approval.
Who it's for
Built for researchers first, patients later.
We start with professionals who already work with ECG and glucose data. Consumer use comes only after clinical validation and regulatory approval.
First customers
Clinical researchers & hospitals
Hospital teams, clinical research organisations and pharma groups studying diabetes and cardiometabolic health, who hold ECG and glucose data and need ECG-based glucose estimates they can analyse and validate.
Next
Digital-health & wearable companies
Companies whose devices or platforms already record ECG and want to explore glucose-related insights for their users.
Long term
People with or at risk of diabetes
Easier, more frequent glucose checks without a finger-prick, but only after clinical validation and regulatory approval.
The product
What goes in, what comes out.
Each step is labelled with where it stands today. Our models exist at research stage; the report layer and the product around them are planned.
01 · In
ECG recording
A 12-lead or single-lead ECG from a clinical machine, bedside monitor, or wearable.
Built (research stage)
02 · Daibeats models
Glucose estimate
Our pipeline extracts 51 heart rate variability and waveform features, and our models estimate glucose with a confidence level, weighting errors in dangerous ranges more heavily.
Built (research stage)
03 · Claude
Readable report
Claude, Anthropic's AI model, turns the estimate, its confidence and the ECG features behind it into a short plain-language report and highlights readings in critical ranges. Claude explains the model's output; it does not make the estimate or a diagnosis.
Planned
04 · Out
For the researcher or clinician
A glucose estimate, its confidence, and a report they can read in a minute, for research and clinical review rather than direct-to-patient use.
Planned
Our story
From research to startup.
Daibeats builds on research into ECG-based glucose estimation that was carried out before the company was founded, as part of Md Basit Azam's doctoral work in the Department of Computer Science & Engineering, Tezpur University.
In 2026, Daibeats was founded as an independent startup to develop this technology into a practical healthcare product. Daibeats is a separate venture and is not a university lab.
Current stage
Research & product validation.
Early-stage · Bootstrapped · Founded 2026. Daibeats is developing and validating its ECG-based glucose estimation technology using clinical and wearable datasets. Daibeats is pre-launch: there are no product users, customers or revenue yet.
01 · Completed on retrospective data
Model development
51-feature ECG pipeline and safety-aware models developed on clinical and wearable datasets.
02 · In progress
Validation
External, subject-independent validation across datasets, reported with clinically meaningful error metrics.
03 · Planned
Product development
A first product prototype for research and clinical partners, including Claude-generated plain-language reports.
Founder
Who is behind Daibeats.
Md Basit Azam
Founder & Lead Researcher, Daibeats
PhD researcher in the Department of Computer Science & Engineering, Tezpur University, working on machine learning for physiological signals. Leads Daibeats' research and product development, from clinical dataset engineering and ECG feature pipelines to safety-aware modelling and external validation.