ML Systems Engineer · FinTech ML

I build production-ready
ML systems for FinTech.
Reliable. Instrumented. Deployed.

From fraud detection pipelines to cloud-native inference APIs — I engineer ML systems that hold up where it matters: in production, under real data, with real consequences.

View audit service See projects
7
Projects shipped
0.880
Fraud PR-AUC
0.921
Risk model AUC
2
Live APIs
AWS+K8s
Cloud stack
Production finding · Project 06 · 13-experiment benchmark
6 of 13 fraud model configs

had Recall@P90 = 0 — meaning at the precision threshold where you'd actually deploy, the model caught nothing — while still posting acceptable F1 scores. Standard metrics hide operationally broken models. This is what production ML reliability work exists to catch.

Work

Production ML portfolio

Seven projects across the full ML systems stack — modelling, deployment, monitoring, and cloud infrastructure.

Cloud · Inference
Cloud-Native Fraud Detection API
Production FastAPI inference service with Registry-first model loading, containerised and deployed to AWS EC2 via ECR. Six Kubernetes manifests written and validated. GitHub Actions CI/CD pipeline.
AWS ECR · EC2 (eu-west-1) · Elastic IP 108.128.140.230 · K8s manifests · Registry-first MLflow loading
FastAPI Docker AWS ECR AWS EC2 Kubernetes MLflow GitHub Actions
↗ GitHub ↗ Live API
ML · FinTech
Credit Card Fraud Detection Pipeline
13-run experiment benchmark across five class imbalance strategies on the ULB dataset. XGBoost + class weighting won: PR-AUC 0.880, Recall@P90 0.837. Central finding: 6 of 13 configs had Recall@P90 = 0 despite acceptable F1.
PR-AUC 0.880 · Recall@P90 0.837 · 13 experiments · MLflow Model Registry · Gradio demo live
XGBoost scikit-learn MLflow DagsHub Gradio pytest
↗ GitHub ↗ Live Demo
MLOps · Monitoring
Production ML Monitoring API
FastAPI monitoring layer on top of the Project 04 heart disease model. Custom pandas-based drift detection, containerised, deployed on Railway with GitHub Actions CI/CD. Seven passing pytest tests.
Deployed on Railway · Custom drift detection · /monitoring/drift · /monitoring/stats · 7 tests passing
FastAPI Docker Railway pandas pytest GitHub Actions
↗ GitHub ↗ Live API
ML · Risk
Heart Disease Risk Pipeline
End-to-end ML pipeline on the UCI Heart Disease dataset (920 rows, 4-hospital combined version). Random Forest best model: AUC 0.921, Recall 0.92. Key finding: GridSearchCV tuning reduced AUC from 0.921 to 0.915 vs baseline.
AUC 0.921 · Recall 0.92 · MLflow on DagsHub · Gradio demo on HF Spaces
scikit-learn Random Forest MLflow DagsHub Gradio HF Spaces
↗ GitHub ↗ Live Demo
Services

Production ML reliability work

Structured audits and implementation work for teams that need their ML systems to hold up in production.

STAGE 01 — AVAILABLE NOW
Production ML Readiness Audit
$250
A focused 5-day audit of your production ML system against a five-point reliability framework. You get a prioritised recommendations report — not a generic checklist.
  • Model evaluation & threshold review
  • Class imbalance & data quality assessment
  • Deployment architecture review
  • Monitoring & drift detection assessment
  • Prioritised recommendations report
Book on Contra →
STAGE 01B — AFTER 3 AUDITS
Fraud Detection Reliability Audit
$500–750
Specialist audit for banks, payment companies, and FinTech startups. Focuses on recall collapse, threshold risk, and operational failure modes that standard benchmarks miss.
  • Recall@precision threshold analysis
  • Imbalance strategy benchmarking
  • Temporal validation review
  • Fraud-specific monitoring gaps
STAGE 02 — IMPLEMENTATION
Production Deployment Package
$2,000–4,000
Full implementation engagement: API hardening, CI/CD setup, cloud deployment, MLflow integration, and model registry. Scoped after initial audit.
  • FastAPI inference service
  • Docker + cloud deployment
  • GitHub Actions CI/CD
  • MLflow model registry setup
  • Monitoring endpoints
STAGE 03 — RETAINER
Production ML Support
$500–2,000/mo
Monthly production ML reliability support: drift reviews, monitoring maintenance, one new feature per month, and direct Slack access.
  • Monthly drift review
  • Monitoring maintenance
  • One new feature/month
  • Direct Slack access
Background

ML Systems Engineer, independent

Building production ML systems since 2024. Focused on the intersection of model reliability, MLOps infrastructure, and FinTech applications.

Certifications
  • GCP Professional ML Engineer
  • Machine Learning Specialisation — Andrew Ng
  • IBM Data Science Professional
  • Machine Learning in Production
  • IBM DevOps & Software Engineering
  • Anthropic: Building with Claude API
Core stack
  • Python · XGBoost · scikit-learn · pandas
  • FastAPI · Docker · Kubernetes
  • AWS ECR · AWS EC2 · GCP
  • MLflow · DagsHub · Gradio
  • GitHub Actions · pytest
  • Hugging Face Spaces
Research focus
  • Concept drift in production fraud detection
  • Class imbalance under temporal validation
  • Metric failure modes under extreme imbalance
  • ML reliability in real-time financial systems
Platforms
Contact

Work with me

If your team is shipping ML models to production and you're not certain how they'll behave under real data, concept drift, or class imbalance — that's exactly where I work. Start with a Production ML Readiness Audit.

↗ Book audit on Contra ↗ LinkedIn ↗ GitHub ↗ Medium