Building AI-first platforms from zero to scale · Financial Inclusion · Mobility · Analytics · Berkeley MBA · IIT Kharagpur
Every LMS ever built was designed for three people: the businessman with liquid assets, the salaried employee, and the old-money landowner. Everyone else — the panwari stocking up before Holi, the kharif farmer on a 2-acre plot, the woman starting a dairy — was simply left out. Not underserved. Structurally excluded.
At Avanti, we built a Loan Markup Language with 88+ dimensions that lets any loan product — daily compounding, variable instalments, pay-when-you-earn for gig workers, hand-written repayment schedules — be configured and launched without an engineer in the loop. Today there are 571 active loan products on the platform, zero engineering involvement. Co-lender integration time dropped from a year to a few weeks. The platform survived COVID because moratorium and restructuring were already built in — most competitors didn't.
The hard tradeoff: we accepted 5 years of slow development before the hockey stick in capability arrived. Platform thinking requires that patience. The lesson I carry: no one can logically explain all future use-cases of a platform from past data. Founder conviction for "the right thing for the ultimate user" is as load-bearing as any roadmap.
80% of Avanti's repayments came in as cash. The borrower paid the agent. The agent's pocket held the money. Avanti had no visibility. The borrower kept accruing interest. The agent could sub-lend it, lose it, or simply walk. Three parties, three different kinds of fraud risk, zero digital trail.
The instinct was to wait for digital adoption. We made a different bet: पानी नीचे को बहता है — water flows downhill. Digitisation was inevitable. We built for where the puck was going. Every loan got a personalised, fraud-proof QR code with the borrower's photo and exact amount. Cash accepted by an agent was recorded on platform instantly, credited to the loan in real time — the borrower stopped accruing interest the moment the agent's hand closed around the note. Agent wallets became virtual accounts with full audit trails. Branch closure became a discipline metric.
The result: the CEO can now say with confidence that ₹1Cr sits across 70 partners nationwide, max ₹2L at any single point, and 95% of collections close within 1 day. The field research lesson that stuck: we were told women in villages don't carry phones. True — their husbands take the phones to the city. But the women agreed to carry a phone on repayment day. People make things up to protect the status quo. Extreme field research is the only antidote.
Consider a cooperative that gives women access to commercial sewing machine workstations. Owning a machine would transform their income. But for the first several months, they train on cardboard, make practice pieces, earn nothing. Then around month six, brands start sending collar and cuff orders. By month twelve, they're cutting full suit-lengths. Their cashflow curve looks nothing like a standard EMI schedule — it starts at zero and ramps slowly. A monthly repayment from day one doesn't burden them. It guarantees they fail.
The insight that drives our hyperflexible LMS: as long as a loan system puts an unfair timing burden on the borrower, it is either predatory or it is about to produce an NPA. It's not a question of whether the borrower will struggle — it's a question of when. The fix is an LMS that supports variable instalments, gradual step-ups, pay-when-you-earn triggers, and automatic switches between aggressive and passive repayment plans — while staying RBI-compliant and co-lender compatible.
The tradeoff is real: most co-lenders want cookie-cutter. Credit bureaus aren't designed for non-standard schedules. This rewrites a hundred years of banking convention, and the GTM will span a decade. Cooperatives and women's organisations have welcomed it. The wave hasn't arrived yet. The highway is built.
Jagriti Yatra put 600 young people on an 18-day train journey across India to ignite the spirit of entrepreneurship. The train had no bathing solution. The obvious answer involved plumbers, carpenters, plywood, pumps, showers — and drilling into railway property. We rejected all of it.
A team of three went back to first principles. Why does a human being need a bathroom? Strip away comfort and luxury and you arrive at something much simpler: privacy. And privacy, at its core, is an optical problem — not a structural one. The second insight: whatever we built had to survive 18 days of a shaking, rattling train. That means flexibility, not rigidity. Rigid structures crack. Flexible ones adapt.
We worked with high-altitude tent makers to bring the concept to life. The solution: a 2ft × 2ft × 6-inch bag that unfolded into a full standing cuboid bathroom — erectable in minutes using a mechanism borrowed from a blind man's walking stick. It had pockets for shampoo, a small flap for a door, and strings that tied directly to the train's existing luggage racks. No nails. No drills. Zero damage to railway property.
Water was carried in by bucket — which actually saved water by making usage conscious. A small tin bathing tub collected the runoff, drained by a PVC pipe laid on the floor to a hole the train already had. Two full coaches — one for men, one for women — were converted. The tents packed back into their little bags at the end. The tin boxes, being metal, were recycled without a second thought.
A decade later, people still talk about those bathrooms. The lesson: the best solutions come from questioning the problem, not the solution. Most people were solving "how do we build a bathroom." We asked "what is a bathroom actually for" — and the answer fit in a bag.
Usually responds within a day