AWS Builder Center

Builder Center

CredVeda AI

CredVeda AI

Breaking the Black Box of Credit Scoring with Explainable AI An inside look at how CredVeda AI combines Scikit-learn, SHAP, and Streamlit to transform opaque credit scoring into transparent decisions and actionable borrower recourse.

CredVeda AI — Built for Kiro Build Challenge

OVERVIEW

CredVeda AI is an explainable credit underwriting and algorithmic recourse platform that transforms opaque credit scoring into a transparent, actionable decision matrix. It empowers everyday retail borrowers to understand the exact financial drivers behind their assigned credit rating while giving loan underwriters debiased, interpretable risk assessments in real time across global currencies.

PROBLEM STATEMENT

Traditional credit scoring acts as an opaque "black box," denying borrowers without explaining the decision or showing how to improve. Lenders also face regulatory scrutiny over hidden demographic biases and lack accessible tools to provide transparent, actionable recourse.

SOLUTION & WHAT WE BUILT

CredVeda AI bridges predictive machine learning and explainable finance through an interactive fintech terminal:
  • Feature Engineering Pipeline: Computes core banking indicators in real time, including Debt-to-Income (DTI) and disposable income surplus, prior to inference.
  • Multi-Currency Global Terminal: Supports seamless toggling between Indian Rupee (₹), USD ($), EUR (€), and GBP (£) with automated scaling, safe monthly EMI limits, and 50% FOIR loan sanction calculations.
  • Interactive SHAP Interpretability: Native Plotly bar charts powered by shap.TreeExplainer visually dissect each parameter's positive pull or risk drag on the assigned tier.
  • Algorithmic Recourse Engine: Automatically simulates step-wise debt reduction to provide applicants with exact capital prepayment targets required to cross into an improved rating tier.
  • Stress-Test Simulation & Audit Dossier: Allows borrowers to test score resilience against macroeconomic shocks (income drops, liability surges) and export an Equal Credit Opportunity audit report as a structured JSON dossier.

HOW WE USED KIRO

CredVeda AI was built using Kiro's Spec-Driven Development (SDD) framework rather than unstructured prompt-and-patch coding:
  • Kiro Specs Engine (.kiro/specs/): Outlined formal acceptance criteria in requirements.md, architectural component boundaries in design.md, and tracked implementation checklists in tasks.md.
  • Steering Directives (.kiro/steering/): Enforced codebase-wide safety standards—such as zero-division guards (1e-5) in financial calculations, deterministic seeds, and strict decoupling of ML artifacts from presentation layers.
  • Automated Agent Hooks (.kiro/hooks/): Implemented validation hooks that run integrity checks on model serialization and verify Python typing consistency across src/ modules.
  • Crew Role Specialization (.kiro/crew/): Partitioned system development across dedicated agent personas (Data/ML Engineer, XAI Recourse Engineer, and Frontend UI Architect) to build data loaders, explainability routines, and dashboard components in parallel.

ARCHITECTURE & TECH STACK

  • Core ML & Data Processing: Python 3, Scikit-learn (Random Forest Classifier), Pandas, NumPy, Joblib
  • Explainable AI (XAI): SHAP (TreeExplainer for local feature contributions)
  • Frontend UI & Visualizations: Streamlit, Plotly Graph Objects (Interactive Gauges, Radar Matrix, SHAP Contribution Bars)
  • Developer Platform & Workflow: AWS Kiro (Specs Engine, Steering Files, Agent Hooks, Crew Architecture)
  • Hosting & Infrastructure: Render Web Service with Custom Domain binding & automated SSL/TLS

LINKS & DEMO

TEAM / BUILDER DETAILS

  • Builder Name(s): Harsh Jain
  • Track: FinTech & Developer Tools / GenAI
Built during the Kiro Build Challenge conducted jointly by AWS Student Builder Group GEU & AWS Student Builder Group PIET.
Challenge Details:https://awssbggeu.com/challenges/kiro-build-challenge 
# kiro# kiro-agent# machine-learning# ai
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article