$76M in originations inside the first year.
Lending Platform
Overview
The client had 250,000 small business customers and no lending product. They wanted to put capital in those customers’ hands, which meant standing up an entire credit business (origination, underwriting, servicing, risk, compliance) from an empty repository. Shikhar was hired to build the engineering function that would do it, and stayed to run it for three and a half years.
The business context
Small businesses are chronically underserved by traditional lenders. They are too small to underwrite manually and too idiosyncratic for consumer credit models. The client already held the thing that makes them underwritable, the transaction history flowing through their business accounts, but none of it was wired into a lending system, because there was no lending system.
Building one is not a feature. It is a regulated business with its own architecture: a decisioning path that has to be explainable to a regulator, a servicing ledger that has to reconcile to the cent, a risk surface that shifts as the portfolio ages, and an audit trail for all of it. Getting any one of those wrong is not a bug. It is an enforcement action.
The mandate was to launch the lending product, and then to keep it safe as it grew.
Key decisions
Credit policy belongs outside the codebase.
The highest leverage architectural decision was to separate credit policy from application code. Underwriting rules, credit limits and rescoring logic went into a dedicated decision engine that the risk team could reason about and change directly, rather than sitting in services only engineers could touch. Policy changes stopped being deploys, and the people accountable for risk stopped filing tickets to adjust their own strategy.
That one boundary is what later made proactive risk management cheap. Portfolio wide rescoring and automatic credit limit adjustments arrived as configuration rather than as projects.
Build the team as deliberately as the system.
A credit platform outlives the people who launch it, so the engineering organisation had to be designed with the same care as the architecture. That meant hiring for ownership rather than throughput, running real performance conversations, and making sure no part of a regulated system had exactly one person who understood it.
Treat compliance and security as architecture.
Sensitive credit data and regulatory obligations were designed into the system boundaries from the start. Access control, retention and traceability became structural properties rather than controls bolted on before an audit.
What was delivered
A working lending product inside a live neobank, launched and then operated for three years across a growing portfolio.
- An origination and underwriting path from application through decision to disbursement, with an explainable decision record behind every outcome.
- A decision engine integration that let the risk team ship policy changes without engineering in the loop.
- Servicing and portfolio systems that stayed reconciled as the book grew.
- An engineering team that ran the platform day to day, with the on call, review and incident practices that keep a regulated system boring.
The product reached $76M in originations within a year of launch, and the credit division became a standing business line rather than an experiment.