LoanLens — Know Before You Sign

the loan lens UI

I## The Moment That Started Everything

I watched someone close to me sign a student loan agreement they didn't fully understand. They were excited — university acceptance is a big deal. Nobody told them to read page 12. Nobody explained what "penalty APR" meant. Nobody flagged the automatic rollover clause buried in the fine print.

Three years later, that clause cost them thousands of dollars they couldn't afford.

What hurt most wasn't the money. It was knowing that if they'd had the right tool — something that could just explain what they were signing — the outcome would have been completely different.

That's why I built LoanLens.

What LoanLens Does

LoanLens is an AI-powered financial agreement translator. Upload any loan document — student loan, car loan, payday contract, lease agreement — and receive a complete breakdown in under 60 seconds:

💰 Real total cost — not just the advertised rate, but everything you will actually pay
🚨 Hidden fee detector — every fee not prominently disclosed, flagged in plain English
⚠️ Penalty clause radar — every trigger that could cost you money, explained clearly
🔒 Rights waiver alert — arbitration clauses and class-action waivers you might unknowingly sign away
📊 Risk grade (A–F) — an instant gut-check before you commit
🔁 Alternative comparison — see how your loan stacks up against current market rates

The Problem Is Bigger Than One Person

Over 60% of the 45 million student loan borrowers in the US alone do not fully understand the terms of the loan they signed.
— Consumer Financial Protection Bureau

The average hidden fee discovered after signing a loan agreement is over $1,200. Predatory lending costs American borrowers $9.5 billion+ annually. And the people who suffer most — first-generation students, immigrants, gig workers, young adults — are exactly the people who can't afford a lawyer to read the document for them.

The ability to understand a financial agreement is currently a privilege of wealth. LoanLens changes that.

How I Built It

LoanLens is built on a lean, purpose-built technical stack:

Layer	Technology
Frontend	React.js — clean, mobile-first UI
PDF Parsing	PyMuPDF / PDF.js for document text extraction
AI Engine	Large language model with a custom financial clause prompt layer
Backend	Python FastAPI
Database	PostgreSQL for anonymized analysis history

The core technical insight is that general AI is not enough. A raw ChatGPT query doesn't know that clause 12 of a standard payday agreement is a rollover trap. LoanLens uses a custom financial clause taxonomy — a structured library of 200+ clause types, red flag patterns, and fee structures — built specifically for this domain. That specificity is the moat.

What I Learned

Building LoanLens taught me that the hardest part of a fintech product isn't the technology — it's the trust. People are sharing sensitive financial documents. Every design decision, from zero data retention on free tier to anonymized storage on premium, was made with that in mind.

I also learned that AI prompting is an engineering discipline in its own right. Getting a language model to produce structured, consistent, legally-careful financial analysis — rather than vague summaries — required dozens of iterations on the prompt layer and clause taxonomy.

Challenges I Faced

1. Accuracy vs. Speed
Financial documents demand precision. Early versions of LoanLens were too vague — they summarized without flagging. Tuning the AI to be specific, actionable, and clause-level accurate without being slow took significant iteration.

2. Scope Creep
Financial agreements come in hundreds of formats — federal student loans, private loans, auto leases, apartment leases, payday contracts. Deciding which to support first (and which to save for later) was harder than expected.

3. The "Just Use ChatGPT" Problem
The biggest challenge was articulating why LoanLens is better than a raw AI query. The answer — structured output, clause taxonomy, risk grading, alternative comparison — had to be demonstrated, not just described. Building the demo was non-negotiable.