L3:credit-scoring-poc · AI Wing · Listed
Explainable credit scoring in the browser
Browser-based gradient-boosted credit scoring that advises, while the credit officer decides.
- Role
- Sole engineer
- Period
- May 2026 to Sep 2026
Results
In-browser
Client-side scoring using ONNX Runtime Web
PoC models were trained on synthetic and public data. No accuracy metric is claimed.
Verified
Gallery
Problem
The credit team needed quick, explainable scoring advice while keeping applicant data out of server-side models.
What I did
- I trained XGBoost and LightGBM models and exported them for ONNX Runtime Web.
- I combined model output with deterministic policy rules and manual-review flags. Scoring remains advisory; the credit officer decides.
Tech stack
- XGBoost
- LightGBM
ONNX Runtime Web
Next.js
L2 · Related career entryTechnical Consultant, Software & AI · Metrodata (PT Mitra Integrasi Informatika)