Data Analytics • Data Engineering • Product Management Case Study
This repository contains a fully deployed, AI-powered Credit Risk Classification Platform designed as a multi-disciplinary case study spanning data analytics, data pipeline design, and product management.
The platform accepts a single-user Excel file containing financial and credit history data, processes it entirely in memory, and instantly classifies the individual into Low / Medium / High credit risk, along with analytical dashboards.
While the application is implemented using Lovable AI and GPT-5.1, the primary focus of this project is on data design, analytical reasoning, pipeline structure, and product decision-making, rather than infrastructure-heavy engineering.
Early-stage credit risk assessment is often:
- Spreadsheet-driven
- Manual and time-consuming
- Inconsistent across analysts and teams
Despite using structured data, many organizations lack a lightweight, standardized mechanism to convert raw financial inputs into fast, interpretable insights during the initial screening phase.
This project explores how a stateless, AI-enabled analytics platform can:
- Reduce time-to-insight
- Standardize first-level credit evaluation
- Preserve privacy by avoiding data persistence
- Fit naturally into existing analyst workflows
- Data analysts performing credit or risk analysis
- Risk and financial consulting teams
- Product managers prototyping AI-driven analytics tools
- Data teams building decision-support systems
The platform accepts a single-row Excel file with the following required columns:
| Column Name | Description |
|---|---|
| Age | Age of the individual |
| Annual Income | Total yearly income |
| Monthly Expenses | Average monthly expenses |
| Existing Loans | Number of active loans |
| Monthly EMI Total | Sum of all EMIs |
| Credit Utilization (%) | Credit card utilization ratio |
| Payment History Score | Proxy score for repayment behavior |
| Past Defaults | Count of historical defaults |
| Credit Inquiries (6M) | Recent credit inquiries |
| Savings Amount | Total liquid savings |
- Exactly one user row is allowed
- All numeric fields are type-checked
- Percentage fields are capped within valid ranges (e.g., utilization 0–100%)
- Missing or invalid values are handled explicitly before analysis
This section demonstrates schema design, validation logic, and real-world data handling.
Each column is mapped to its business meaning to ensure interpretability for non-technical stakeholders.
Examples:
- Payment History Score reflects repayment consistency
- Credit Utilization indicates reliance on revolving credit
- Savings Amount represents a financial safety buffer
The platform follows a stateless, in-memory data pipeline, where each upload is processed independently:
- Excel file ingestion
- Schema validation and row enforcement
- Data cleaning and normalization
- Feature derivation (DTI, EMI burden, utilization)
- Risk classification
- Dashboard metric generation
- UI rendering
No intermediate or final data is stored.
- Stateless and idempotent processing
- Privacy-first design
- Immediate error feedback on invalid inputs
This highlights data engineering principles without infrastructure complexity.
- Debt-to-Income (DTI) Ratio
- Monthly EMI Burden
- Credit Utilization Percentage
- Payment History Score
- Savings Buffer
Each KPI is derived using standard financial logic and is designed to be:
- Interpretable
- Comparable
- Suitable for dashboard visualization
- Income vs Expense comparison
- EMI burden distribution
- Credit utilization overview
- Savings comparison
All insights are descriptive and factual.
No recommendations or financial advice are generated.
Key product decisions made during development include:
- Excel-based input to align with existing analyst workflows
- Stateless architecture to minimize privacy and compliance risks
- Risk bands instead of numeric scores for better interpretability
- Dashboard-first output for executive and stakeholder readability
- No persistence or history to keep the system lightweight and demo-ready
These trade-offs were intentionally chosen to balance speed, usability, and clarity.
- Lovable AI — Used for end-to-end application orchestration and UI generation. Lovable internally generates a modern web frontend using standard web technologies (HTML, CSS, JavaScript) and component-based, React-style architectures, while abstracting low-level implementation details from the user.
- Frontend Stack (abstracted by Lovable AI) — Responsive, browser-based UI built using HTML/CSS, JavaScript, and state-driven, component-based rendering patterns similar to React, enabling dynamic dashboards and instant re-rendering on file upload.
- OpenAI GPT-5.1 — Core analytical engine used for in-memory data preprocessing, credit risk classification (Low / Medium / High), KPI computation, and generation of chart-ready dashboard data in a fully stateless processing pipeline.
The full implementation code is included in this repository.
- Batch uploads for portfolio-level analysis
- Configurable risk thresholds
- Explainability and audit layers
- Integration with BI and reporting tools
A deployed version of the platform is available here:
https://gleam-credit-scan.lovable.app
This project is a prototype and learning exercise.
Outputs are illustrative and not intended for use in real-world credit decision-making.