Inside MARG: Building an AI Loan Advisor for India

MARG exists to make home ownership simpler for aspiring Indian families. But talk to their support team for an hour and you will hear the same friction point over and over: most first-time borrowers do not understand what they qualify for, what EMI actually means for their monthly budget, or why one lender offers a different rate than another.

That is the exact problem a group of Academy builders picked up as their capstone challenge: build an AI assistant that could explain loan eligibility, EMI, and lender options in plain language, for people who have never taken a loan before.

Starting with the real constraint, not the fun part

It is tempting to jump straight into prompt engineering and model selection. Instead, the first two weeks were spent with MARG's product team understanding who the actual user is: often someone applying for their family's first home loan, sometimes in a second language, usually anxious about getting it wrong.

That context changed the entire approach. The assistant needed to ask clarifying questions instead of assuming financial literacy, explain its reasoning instead of just outputting a number, and flag when a user should talk to a human loan officer instead of trusting the AI blindly.

What got built

• A conversational assistant that walks users through eligibility using their income, existing obligations, and credit history.
• An EMI calculator that explains trade-offs in plain language, such as shorter tenure versus lower monthly payment, instead of just showing a table.
• A lender comparison layer that surfaces the two or three best-fit options instead of overwhelming users with every possibility.
• Guardrails that hand off to a human loan officer whenever the assistant's confidence drops below a set threshold.

The prototype now runs as part of MARG's actual product testing, with real (anonymized) usage data feeding back into the next cohort's work. That loop, learners building for a real company, and that company's real users shaping what gets built next, is exactly the kind of feedback tutorials cannot replicate.

If you want to work on a challenge like this yourself, the AI for Developers & Builders and Finance with AI paths both feed directly into MARG's active challenges on the Companies page.