Suraj Gharde

All projects

CreditIQ AI

AI-Powered Credit Risk & Fraud Detection System

Code on GitHub(opens in new tab)

Key numbers

  • 14 financial line items extracted by hybrid OCR

How it works

  1. 1Input

    MSME loan documents

  2. 2Extract

    Hybrid OCR (Tesseract + AWS Textract) pulls 14 financial line items and flags GST mismatches

  3. 3Score

    XGBoost default-risk model with SHAP explanations; graph analysis traces circular trading

  4. 4Output

    Early-warning dashboard and LLM-generated Credit Appraisal Memo PDFs

What I built

  • Engineered a credit-appraisal pipeline for MSME loans with hybrid OCR (Tesseract + AWS Textract) that extracts 14 financial line items and flags GST mismatches.
  • Modeled default risk using XGBoost with SHAP explanations and traced circular trading via graph analysis, feeding an early-warning dashboard and LLM-generated Credit Appraisal Memo PDFs.

Stack

Python, FastAPI, React.js, XGBoost, NetworkX, Cohere LLM

Tesseract, AWS Textract, SHAP