By Kangana Pandiya, PGP 2022
In April this year, we sat across from a senior official at a state government audit department that had just taken delivery of a cluster of new GPUs, a hardware order more commonly associated with a frontier AI lab than a government audit office. The GPUs were arriving faster than anyone locally had a plan for them. The department’s mandate covered welfare-sector audit: schools, hospitals, and social-welfare schemes, each generating years of scheme documents, utilisation certificates, and audit paragraphs that no human team could cross-reference at scale. They wanted a system that would run entirely on their own machines, on their own data, inside their own walls, with everything staying in-house.
That distinction, between renting intelligence and owning it, has stayed with us. We both joined McKinsey after IIM Ahmedabad. Soumya then spent time at a geospatial AI startup before we started Livo Assistant together. Across that audit department and the institutions we describe below, we kept running into the same underlying problem wearing different paperwork: records too messy and too inconsistent to compare against each other, with discrepancies buried inside them that nobody had time to find. Livo Assistant’s focus has narrowed to exactly that: building AI that standardizes unstructured records into clean, comparable data, then surfaces the anomalies hiding inside them. We’re now working with some of India’s largest data providers to build this out at scale.
The same idea shows up at a completely different scale in rural Rajasthan, where the Centre for Microfinance works with roughly four lakh self-help groups and three lakh volunteer bookkeepers, who have historically tracked attendance, savings, and loan repayments in paper registers. The entire bookkeeping system now lives inside WhatsApp, a platform every bookkeeper already has open all day. The tool stayed simple. We changed who it was built for. A ledger entry that used to take a volunteer twenty minutes of handwriting, cross-checking, and eventual re-entry into a district office’s spreadsheet now takes under a minute; it’s also logged the moment the meeting happens, already in a standard, comparable format. We’re currently scaling this from fifty groups to a thousand, and the request from CMF was simply to make sure this keeps working when fifty times as many people are using it at once.
We see a version of this same story inside chartered accountancy firms and mid-sized manufacturers, who make up much of India’s actual services and industrial base, far from the venture-funded end of the economy. One CA firm we work with put it bluntly: “The volume of documents we were processing by hand every month was unsustainable.” It’s a firm asking for the same paperwork to take less of its partners’ week.
These stories rarely make the news. A government audit office running its own GPUs and a WhatsApp thread in a Rajasthan village share a premise: the infrastructure of Viksit Bharat includes the invisible back office of the country, audit trails, ledgers, compliance filings, the paperwork that every functioning institution runs on and that has to be standardized before anyone can tell what’s wrong with it.
We started Livo Assistant believing that the more interesting AI problem in India was building systems patient enough to sit inside a government department’s data-sovereignty rules, or a self-help group’s WhatsApp habits, standardize what they produce, and flag what doesn’t add up, without either institution ever having to think about the AI underneath. Viksit Bharat @2047 will be built inside institutions as much as by individuals. We suspect it will look a lot like this: unglamorous, largely invisible, and already underway in a hundred back offices across the country.
Author Bio:
Kangana Pandiya and Soumya Sharma are co-founders of Livo Assistant (livoassistant.com), an AI company building data standardization and anomaly detection tools for India’s government, financial services, and enterprise back offices.
