DataNest Ventures

When COVID-19 struck in early 2020, VIT brought the summer break forward and the free months turned into an opportunity. I carried on earlier work on social network analysis — how people actually cross paths, and what the shape of those encounters says about what travels through them. The research gained momentum, became CoviFight, and CoviFight became the foundation of DataNest Ventures, founded by three of us.

The product was a hybrid mobile and web application in Java, Python, Node.js and Gremlin Graph DB, running on Kubernetes on Azure. The premise was narrow. A tracing app is only useful if enough people install it, and in 2020 the thing standing in the way of installing one was being asked to surrender an identity to whoever ran it.

Every user held a decentralised verifiable ID on a blockchain, with their public key as the sole identifier. No name, phone number, email address or device identity was ever written to chain. Nothing about a person was stored anywhere that could be used to identify them, which was a property of the design rather than a policy on top of it.

Capture was ordinary Bluetooth work. Devices ran continuous BLE scanning, and when two crossed paths each stored the other's ID locally. Those IDs resolved to public keys already on chain, so an encounter could be verified later without either party being named.

Because encounters resolved to public keys and BLE IDs that were already on chain, a device did not need a server to work out its own risk. It read what was public, computed locally, and published one number back. Federated analytics at the edge: every device computed its own risk locally and only anonymous scores were ever shared — never a contact history, and never model weights, because no shared model was being trained.

The multiple is the headline. The property underneath it is the part I would defend first, and it is the answer to the obvious question — why a blockchain rather than a database. Because the data was on chain, every edge node could perform the analytics independently and verify the result independently. A light client could check the epidemiology for itself instead of trusting an operator's output.

Neither a central server nor a conventional distributed database gives every participant that. Without it, the anonymity guarantee is a promise rather than a property, and a promise is a poor thing to ask a country to install.

Nothing was ever compulsorily uploaded. A device could optionally send its connected-device list to a central server sitting outside the blockchain, and whether that happened at all depended on a country's regulations and on how deep its government wanted the analytics to go. Where a jurisdiction opted in, the data collected was still fully anonymous, and it fed time-series analytics alongside graph and social-network analytics — linking encounters between anonymous identities and estimating infection probability against continuously updating variables.

Consent and data residency were configuration rather than a rewrite. That is what made the system deployable across European jurisdictions with very different privacy regimes, and it is what made the conversations with them possible at all: a government could evaluate the mode that fitted its own law rather than the one we happened to prefer.

I ran the company and I led the engineering, and at around twenty-one those were one job rather than two. Twelve developers, a compressed 2020 timeline under lockdown conditions, and Scrum Master duty across cross-functional teams — while contributing the machine learning models and the database schema myself. I set the technical direction: Gremlin Graph DB and time-series analytics to find the interaction patterns, an integrated blockchain layer for privacy and verifiability, and a Kubernetes deployment on Azure sized to collect and analyse terabytes of human interaction data.

The outward half was mine too. We secured a grant from The Meta Group, a European investment group, along with $6,500 in AWS credits and $1,875 in Wolfram credits. I met more than fifty people across Europe about how CoviFight could raise COVID-19 detection — public officials, academic institutions, corporate partners and venture capitalists. The opt-in architecture was what made those meetings go anywhere.

The two that mattered: CoviFight placed 2nd in the Real-time Communication and Prevention challenge at EU vs Virus, the European Commission's pan-European hackathon led by the European Innovation Council, out of 9,000+ participants and 2,000+ teams, winning €1,000. In April, India's Ministry of Human Resource Development named the team among the Top 15 Student Innovators in the country in the SAMADHAN challenge.

Three more followed in the same stretch: The Wolfram Award from Wolfram Research for open-source work using AI or blockchain against COVID-19, 2nd position in Boston University's Resiliency Challenge, and a top-six finish in the Crisis Response track of The Global Hack. All five are credited to DataNest Ventures.

What I took from it was not the awards. It was the habit of asking what a system proves rather than what it claims — and finding out how much of building a company is conversations with people who will never read the architecture, and who have to be given a reason to trust it anyway.