Nirlep Adhikari

We built an assistant that let finance brokers ask questions about lender credit policies and get answers drawn from the policy documents themselves, using generative AI and Amazon Bedrock with a retrieval-augmented generation architecture. AWS published the work as a case study, which I co-authored with two of their solutions architects. It ran in production against real policy documents, where a wrong answer has consequences, so answers pointed back to the source and the broker stayed the decision maker.

I joined Australia's first AI-powered loan comparison and broking platform when the engineering team was small and the product was early, and led it through the work that turns an idea into a system a regulated business can run on. I owned the platform architecture, the AI matching and decisioning layer, the engineering and DevOps functions, and the security and compliance posture behind sensitive financial data across the lender network. During my time the platform was named FBAA Industry Supplier of the Year and recognised on Deloitte's Technology Fast 50.

I work with founders in finance, lending, insolvency and professional services across Australia on the technology decisions that are hard to reverse: platform architecture, vendor selection, engineering hiring, build-versus-buy, and whether an AI proposal will survive production. Every engagement starts with a paid diagnostic rather than a quote, and I take a small number of clients at a time so the work stays hands-on. Some engagements lead to a significant build. Some lead to the advice to build nothing. Both are the job.

I designed and built the cloud infrastructure the platform ran on: serverless architecture on AWS, infrastructure-as-code with CDK, and the CI/CD and DevOps practice the engineering team worked inside. This was hands-on work rather than oversight. The same infrastructure carried a regulated lending platform handling sensitive financial data under Australian privacy obligations, so the security posture, the deployment process and the failure modes all had to hold up in production every day.

Small Business Restructuring can be used only once every seven years, so a director who picks the wrong adviser loses their one shot. ASIC publishes the outcome data, but it sat in raw registry files no business owner would ever open, which meant marketing spend rather than track record won the work. We built a free public platform that ingests that data, normalises it across industries, and presents practitioner track records in a format anyone can read in a minute. No login, no funnel. The engineering was the smaller problem. The harder part was presenting public data fairly.