VyomaResearch
Health AI that is accurate, accessible, and safe.
Our research teams study diabetes, AI in healthcare, and how intelligent systems can support patients between doctor visits.
Clinical
01Turns dense medical literature on symptoms, ranges and complications into plain-English guidance.
AI Methods
02How language models, retrieval and on-device computation can personalise health support without compromising privacy.
Lifestyle & Behaviour
03Food, movement, sleep, stress — the habits no medication touches, and how they compound.
Public Health
04India-specific epidemiology and screening: reaching the 130 million Indians with prediabetes who don’t yet know it.
Safety & Ethics
05Medically conservative, refusal-trained for diagnosis and prescription, privacy-preserving by design.
All research
40 entries · showing 39
Reading your HbA1c — what the number means
A 3-month blood-sugar report card: the ranges, the conversion to average mg/dL, and the limits of A1c when glycemic variability is high.
Why India has so much diabetes
Visceral fat at low BMI, dietary shift, urbanisation, and genetic predisposition: the multi-layered story of why diabetes prevalence in India is the highest per capita in the world.
How DiaCare AI was built
A wrapper-model architecture for a health assistant: Helix 1, a Cloudflare Worker, a 6,500-character system prompt, and local-first health profiles. Engineering notes.
Recognising hypoglycemia early
Eight warning signs, the 15-15 rescue rule, and when severe hypo becomes a 112-call situation. The single most actionable safety topic for insulin users.
Indian meals and portion sizes for diabetes
The plate method, applied to roti, dal, rice, and sabzi. Portions in katoris and palm-fistfuls — practical, India-specific, immediately usable.
Exercise and diabetes — the cheapest medicine
30 minutes a day, five days a week, lowers HbA1c by 0.6-0.7%. The post-meal walk hack alone can drop post-meal sugar by 20-30 mg/dL.
Prediabetes — the warning your body is sending
The 130 million Indians who sit at HbA1c 5.7-6.4%, and the lifestyle interventions that flip 58% of them back to normal within five years.
Gestational diabetes basics
Why pregnancy hormones drive insulin resistance, the DIPSI screening test, and the lifetime Type 2 risk that follows GDM.
Diabetes and your heart
Two of every three diabetics die from a heart event. Why blood vessel damage compounds over decades and what actually protects the cardiovascular system.
Building a diabetes daily routine
Morning, mid-day, and evening rituals that quietly do the heavy lifting. The structure that makes diabetes manageable without consuming your life.
Sleep, stress, and blood sugar
A single 5-hour night raises next-day insulin resistance by 25-30%. Chronic stress keeps cortisol high. The two drivers most plans skip.
Hydration and diabetes
How high sugar dehydrates, how dehydration concentrates sugar, and the daily water target tuned to body weight.
Diabetes medications — a plain-English overview
Metformin, sulfonylureas, DPP-4 inhibitors, GLP-1 agonists, SGLT2 inhibitors, insulin: what each class does and the trade-offs.
Foot care for diabetics
Why neuropathy plus reduced circulation creates the diabetic foot ulcer pattern, and a 5-minute daily check that prevents most of it.
Blood sugar monitoring — when, how, what to look for
Frequency by treatment regimen, target ranges, and how to log readings so patterns emerge in two to three weeks.
10 diabetes myths, busted
Sugar does not directly cause diabetes. Karela does not cure it. Insulin is not a sign you failed. Ten everyday myths and the evidence.
When to see a doctor — not later, now
Six escalation triggers DiaCare AI is trained to flag immediately. Newly suspected diabetes, persistent high readings, recurring lows, non-healing wounds, pregnancy, emergency signs.
Asking DiaCare AI good questions
The patterns that get personalised, actionable answers vs generic ones. How profile context plus specific framing changes what the model can produce.
Powered by Helix 1 — the wrapper model that drives DiaCare
Why wrapping a high-quality foundation model in a domain-specific system prompt, memory layer, and safety overlay beats training a custom medical model for a small team.
IDRS scoring and South Asian risk stratification
The Indian Diabetes Risk Score and its predictive value across urban and rural cohorts. Where DiaCare's risk scanner adapts the standard IDRS frame.
Vision-based screening for diabetic retinopathy
Open question: can a smartphone camera + an on-device vision model flag early DR? Survey of published approaches, accuracy claims, and what remains a referral path to an ophthalmologist.
Federated learning for sensitive health data
Training models without centralising patient data. A primer on horizontal vs vertical federation and what it would take to make federated training viable for Indian diabetes cohorts.
Privacy-preserving health assistants
Why DiaCare keeps the entire health profile in browser localStorage, what the threat model looks like, and where on-device inference is heading.
Glycemic index of common Indian foods
Measured GI values for rice variants, millets, dal preparations, breakfast staples (idli, dosa, paratha), and how cooking method shifts the number.
Continuous glucose monitoring — what the trace teaches
CGM reveals glycemic variability that point-of-care meters miss. Time-in-range targets, dawn phenomenon detection, and how the trace changes coaching.
GLP-1 agonists — mechanism and outcomes
How semaglutide and tirzepatide combine A1c reduction with weight loss and cardiovascular protection. Where the class fits in Indian clinical practice today.
SGLT2 inhibitors — heart and kidney effects
A class designed to flush glucose through the kidneys, with side benefits that may matter more than the original use case. The cardiovascular and renal trial evidence.
Plant-based diets and HbA1c
Meta-analysis-level evidence on vegetarian and vegan eating patterns vs Type 2 outcomes, with caveats for Indian protein adequacy and B12.
Intermittent fasting — safety in diabetes
When extended fasting helps insulin sensitivity, when it triggers dangerous hypoglycemia on insulin or sulfonylureas, and how to approach Ramadan or ekadashi safely.
Family history as a diabetes predictor
How parental Type 2 doubles risk in offspring, the gene-environment interaction model, and the case for cascade screening in high-risk families.
Childhood obesity and Type 2 diabetes in India
The pediatric Type 2 epidemic that did not exist a generation ago. Urbanisation, screen time, and school food environments as drivers.
PCOS and the diabetes connection
Polycystic Ovary Syndrome shares the insulin resistance axis with Type 2. Why early metabolic screening in women with PCOS pays off.
Statin therapy in diabetics — when and why
Current guidelines move toward statins for most diabetics over 40. The cardiovascular risk math, dose intensity choices, and the muscle-pain side effect reality.
Voice-first health interfaces for low-literacy users
Open research: speech-to-intent for Indian-language health queries, accent robustness, and where voice can lower the threshold for elderly and low-literacy patients.
Memory-augmented health assistants
How user-owned health profiles plus retrieval-augmented context can deliver personalised guidance without surveillance. Architecture trade-offs.
Medication adherence patterns
Why diabetes medication adherence drops to ~50% by year three for many patients. Reminder mechanisms, side-effect tolerance, and the role of conversation.
Alcohol and glucose control
Why alcohol can cause delayed hypoglycemia in diabetics on insulin, the carbohydrate content of common drinks, and the case for abstention vs moderation.
Smoking cessation outcomes in diabetics
Smoking accelerates every diabetes complication. Cessation outcomes within 12 months — cardiovascular, microvascular, glycemic — and which interventions work in the Indian context.
AI coaching for chronic disease — what is plausible
No guarantees, but a plausible roadmap: AI as an always-available coach between doctor visits, summarising patterns, surfacing concerns, escalating when needed. Where current evidence supports it and where it does not.
Put the research to work.
DiaCare AI turns this work into a free, friendly companion — in English, Hindi, and Hinglish.