Skip to work
ASK.

Home / Work / meridian

Risk Intelligence · Graph ML · MLOps2026

Meridian

Supply Chain Risk Intelligence Platform

EngineeringProductResearch

Meridian

Problem

Supply-chain disruption signals arrive as multi-source event noise. Operators need scored risk, graph context, and explanations — not another offline notebook forecast.

Approach

Built a real-time risk platform combining Kafka event streams, Neo4j graph retrieval, XGBoost scoring, and authenticated FastAPI services across multi-source supplier and disruption data.

Tracked XGBoost experiments with MLflow, added SHAP explanations and Monte Carlo disruption simulation.

Automated unit and integration testing through GitHub Actions to improve interpretability, repeatability, and release quality.

Results

Kafka + Neo4j · XGBoost + SHAP · MLflow

  • Operational risk scoring over live multi-source supplier and disruption data
  • Explainable XGBoost outputs with SHAP and simulation support
  • CI-backed experiment tracking and release validation via MLflow and GitHub Actions

Stack

KafkaNeo4jXGBoostMLflowFastAPI

What I learned

Risk products earn trust when scores are explainable and the data path is observable. Graph context plus SHAP beats a black-box probability every time.

View on GitHub ↗