CreditWise AI: Building a Privacy-First Loan Advisory Engine for Freelance Brokers with Gemma 2

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

CreditWise AI: Automated Loan Decisioning & AI Risk Advisory Engine

Demo

https://creditwise-loan-system-cxudnrbbqzwh8qlmepmzxh.streamlit.app/

Code

https://github.com/JATIN-PANDAY/CreditWise-Loan-System.git

How I Built It

I built CreditWise AI, an automated loan eligibility prediction system paired with an open-weight AI financial advisory layer designed for micro-finance consultants, freelance loan brokers, and small credit agencies.

Why Does Open Innovation Matter?

Open innovation and open-weight models like Google Gemma 2 are not just technical choices for CreditWise AI—they are essential to making this side-hustle business model viable, secure, and compliant.

Here is why an open-source approach outperforms closed, proprietary APIs for this project:

1. Strict Financial Privacy & Compliance:
   Loan applications contain highly sensitive data—applicant income, credit history, marital status, and loan amounts. Passing this private data through proprietary closed APIs creates significant compliance and data-retention risks, as commercial platforms may log or train on user inputs. With open-weight models, the entire inference pipeline can run locally or on a private server, keeping financial records completely offline and strictly confidential.

2. Predictable Unit Economics for Small Businesses:
   For a micro-finance consultant or freelance loan broker, software margins matter. Closed APIs enforce per-token pricing and unpredictable price increases that can quickly wipe out the profitability of a small SaaS tool. Open-weight models allow us to host inference on fixed-cost infrastructure (like a local GPU or a low-cost DigitalOcean Droplet) with zero per-query API fees, keeping running costs near zero as usage scales.

3. Complete Control Over Model Behavior & Fine-Tuning:
   Closed AI services arbitrarily update or deprecate models, altering prompt responses without warning. Open innovation gives us full ownership over the model weights. We can pin specific versions, fine-tune Gemma on region-specific loan guidelines (like Indian micro-finance or MSME lending rules), and enforce strict JSON schemas without relying on third-party policy changes.

My Agent Session

• Development Log & Agent Session: Session workflow tracked during setup and testing. (Optional: link your DevRelay trace here if recorded).

My First User

I tested CreditWise AI with a classmate acting as a micro-finance loan consultant reviewing a high-risk applicant profile (zero credit history, high requested loan amount).

When the system flagged the application as "Rejected", the Gemma-powered open-source advisor recommended lowering the loan term and adding a co-applicant to lower lender risk exposure.

User Feedback: "Most financial software gives a binary 'Approved' or 'Rejected' without explanation. Having the open AI model analyze the exact metrics and suggest lowering the loan amount so the client can actually get approved makes this a genuine time-saver for client consultations."

They mentioned they would willingly pay a ₹499/month subscription if integrated directly into a WhatsApp bot for instant client pre-screening during field calls.

Prize Categories

• Best Use of Gemma (Integrates Google's open-weight Gemma-2-9B model to power the automated AI financial risk advisory engine)

Thank you to DEV, Hacktoberfest, and the Google Gemma team for hosting this challenge! Open innovation is making AI accessible, private, and practical for real-world side hustles.