FINTECH · ML / RAG · 2026 — en cours
Fvundi
An explainable credit scoring engine tailored to emerging economies, decoupling predictions from natural-language explanations.
- ROLE
- Founder & CTO
- DOMAIN
- FINTECH · ML / RAG
The problem
How to assess creditworthiness from Mobile Money telemetry while keeping every financial decision accountable, transparent, and auditable?
The solution & architectural choices
XGBoost computes the score. TreeSHAP extracts key decision drivers. An ensemble RAG layer — LlamaIndex, Cohere, pgvector, and Claude — synthesizes grounded explanations. The MLOps pipeline features drift monitoring and anomaly tracking.
Key results & outcomes
Uncovered demographic explanation faithfulness disparities despite identical ML predictions, inspiring research on constitutional oversight for hybrid AI.
Technologies & core concepts
XGBoost · LlamaIndex · SHAP · FastAPI