The verification layer for what people consume

We verify everything a human consumes. Convolity AI Private Limited

People take prescription drugs, supplements, herbs and foods together every day, and nobody can reliably tell them whether those things help each other, cancel each other out, or quietly cause harm. We map roughly 4,000 substances into millions of source-cited relationships so any app can answer that question.

Every claim in our data carries a resolvable primary-source ID and the exact verbatim sentence that supports it. If the source does not literally back the claim, we store an honest blank instead of a guess. Our work is at research and development stage.

How it works One engine, three markets
Substances verified
1,194across 34 scientific dimensions
Primary-source links
~33kclickable citations
Fabricated claims
0enforced in code, not prompts
Target coverage
~4,000substances · ~16M relationships

The problem

The information exists. It is scattered, and AI invents the rest.

The facts needed to answer "is this safe, and is it working?" are spread across FDA labels, PubMed, PubChem and clinical registries. When a general-purpose model stitches them together, it produces a fluent, plausible, cited-looking answer that is sometimes simply invented. For a safety question, a confident wrong answer is worse than no answer.

People combine things

Most people take several consumables at once — a prescription, an over-the-counter medicine, a couple of supplements, coffee, alcohol. The combined effect is where the risk lives, and it is exactly what nobody checks.

A lot of it is wasted

Substances cancel each other, duplicate each other, or are already covered by a person's diet. The result is money spent on things that are not doing what the label promises.

Generic AI hallucinates

Peer-reviewed evaluation shows language models fabricate and omit clinical facts, and detect medical hallucinations worse than humans do. Grounding, not fluency, is what a health answer needs.

How it works

Three layers, not a chatbot on top of medical text

We did not build a chatbot over a pile of documents. We built a data layer that cannot hold a guess, a reasoning engine that works from biology rather than lookup tables, and a personalisation layer that makes the answer belong to one specific person.

Evidence-bound knowledge base

A database that cannot contain a guess

A generator model reads primary sources only — FDA drug labels via openFDA and DailyMed, peer-reviewed literature via PubMed, chemistry via PubChem, trials via ClinicalTrials.gov — and proposes candidate facts.

A verifier model from a different vendor's family independently checks each one against the retrieved text, so no single model's bias goes unchallenged. A deterministic code gate then refuses to save any fact that lacks a resolvable source ID and the exact verbatim sentence supporting it. Rejected facts are stored as honest blanks. This is enforced in code, so it cannot drift.

Mechanistic reasoning engine

Reasons about biology instead of looking things up

For each substance we store its actual mechanical properties: which liver enzymes it blocks, speeds up or is broken down by; which physiological pathways it pushes and in which direction; what it competes with for absorption; how long it stays in the body; which nutrients it depletes.

Sixty-one coded rules run over those properties, so we can flag a dangerous or wasteful combination that has never been documented anywhere — because we derive it from mechanism rather than retrieve it from a table. Every substance we add strengthens all sixty-one rules at once.

Personalisation

The answer belongs to one specific person

A guided intake builds a real model of the individual — their goal, age, diet, coffee, alcohol and smoking habits, conditions, allergies, budget, and everything they actually take. The engine runs on that person.

It checks the whole stack together rather than pair by pair, catches substances that cancel each other, builds and re-adapts a daily schedule as items are logged, warns when a medicine is draining a nutrient, and tells people what to stop buying. Every screen carries an evidence grade and a tappable primary-source citation.

Why it is infrastructure

One verified engine, three markets

The same verified engine serves three markets from a single build. The data layer is the product; the interfaces are how different customers reach it.

Consumer product

A person-facing app that reads labels and food from a photo, understands a spoken stack, checks it end to end, and answers the real question: is this working, or am I wasting my money?

Interaction-checking API

An interaction and safety API for telehealth, pharmacy and nutrition platforms — spanning food, herbs and supplements, not drugs alone. The only free public drug-interaction API was discontinued in January 2024; we intend to fill that gap properly.

Grounding corpus for AI

A grounding and evaluation corpus for AI teams that need their health answers to be provably correct rather than merely fluent — every claim carrying its source and verbatim supporting quote.

Why it compounds

Each substance we add creates thousands of new relationships instantly, and the relationship count grows with the square of the substance count. Model inference is becoming a commodity; the scarce asset is trustworthy ground truth — the one thing that cannot be shortcut, and the layer we own.

About the company

Built independently, grounded in public research data

Convolity AI Private Limited was incorporated in India in January 2026 as a private company limited by shares, and is recognised as a startup by the Department for Promotion of Industry and Internal Trade (DPIIT). Our verification pipeline runs against public research institutions' open data programs.

Public open data

US National Library of Medicine (PubMed E-utilities, DailyMed), the FDA (openFDA), NCBI PubChem, ClinicalTrials.gov, and the NIH Office of Dietary Supplements.

Published methods

Retrieval-augmented generation, LLM-as-judge evaluation, and self-consistency decoding — built on peer-reviewed research rather than proprietary black boxes.

Cost that falls with scale

We are moving the bulk of verification onto self-hosted open models, with a fine-tuned biomedical checker, which is what makes full coverage economically reachable.

Research and development stage

Our systems are under active development. They are intended to support the judgement of qualified professionals, not to replace it. None of our software is approved or cleared as a medical device by any regulator, including the Central Drugs Standard Control Organisation (India) or the U.S. Food and Drug Administration, and none of it diagnoses, treats, cures or prevents disease.

Our product

SuppliAi — the verified engine, in a product

SuppliAi is our product, developed and operated by Convolity AI Private Limited. It puts the verification layer and reasoning engine in front of a person, and it is reachable today.

SuppliAi

Available

Clinical decision support for supplement, vitamin and prescription medication interactions. It reads a person's whole stack, checks it end to end against our verified corpus, and produces structured, research-referenced analysis — every claim carrying its evidence grade and a tappable primary-source citation — that can be shared with a clinician or pharmacist.

Visit SuppliAi

People

Leadership

The board of Convolity AI Private Limited, and the people leading engineering and research.

Isarar Siddique

Director · Co-Founder

Leads software and AI systems architecture, including the verification pipeline and reasoning engine.

DIN 10932118

Arshi Niyaz

Director

Director on the board of Convolity AI Private Limited since incorporation.

DIN 10932117

Raiyan Siddique

Co-Founder · Research Lead

Leads clinical research and the evidence curation that underpins the verified corpus.

Get in touch

Contact Convolity

For partnerships, API access, clinical collaboration, procurement or general enquiries.

Telephone
+91 74088 71801
Registered office
29 Motipur, Mehanharhangpur, Patherdewa,
Deoria, Uttar Pradesh 274404, India