14+ years of platform engineering and 2+ years shipping production LLM and agentic systems — LangGraph supervisor-worker agents on AWS Bedrock, RAG pipelines on Vertex AI. Specializing in cloud-native infrastructure on OpenShift and Kubernetes. Based in Toronto.
I spent a decade leading technology and DevOps at Socketworks in Lagos, then moved to Canada and into enterprise platform engineering — multi-tenant OpenShift estates, Istio service mesh, GitLab CI/CD, and the operational constraints that come with serving carriers and financial institutions at scale.
Over the last two years my work has tilted toward AI. I build and ship production LLM and agentic systems — LangGraph supervisor-worker agents on AWS Bedrock, RAG pipelines on Vertex AI, evaluated with RAGAS. More recently I've been building these agents as MCP servers — a Kubernetes cost-anomaly agent and a CI-triage agent that plug directly into tools like Claude Desktop — because the interesting problem isn't just “can the model reason about this,” it's “can it act on live infrastructure safely.” The through-line is full-stack ownership: from model integration to CI/CD, observability, and security.
The side projects are where the AI work started. African Stores Canada is an agentic system that catalogues African businesses across Canada; igbo-rag is quieter — my wife and I both carry Igbo heritage, and I am building a retrieval-grounded translator to help preserve the language for our children.
When I am not in a terminal I shoot photographs on a Sony A7III. It keeps me looking at the world a little differently.
On the A7III — the reason the Igbo project matters.
Certifications
IBM Generative & Agentic AI Developer
2026
HashiCorp Terraform Associate (003)
2024
Google Associate Cloud Engineer
2024
Gov. of Canada Reliability Status
Active
Education
M.Sc. Information Systems Management
University of Liverpool
B.Eng. Electronic Engineering
University of Nigeria, Nsukka
Skills
The toolkit
AI / Agentic Engineering
LangGraph
LangChain
MCP
AWS Bedrock
Vertex AI RAG
Gemini 2.0 Flash
Claude Haiku 4.5
Anthropic API
RAGAS
FastAPI
FAISS
SageMaker
Container & Orchestration
OpenShift 4.x
Kubernetes
Docker
Podman
Helm
Istio
Cloud & On-Prem
AWS
GCP
Azure
OpenStack
IaC & Automation
Terraform
Ansible
AWS CDK
CloudFormation
OIDC
CI/CD & GitOps
GitLab CI
GitHub Actions
Jenkins
ArgoCD
Azure DevOps
Security & Quality
Checkmarx
Mend
CodeQL
Trivy
SonarCloud
Bandit
Observability
Prometheus
Grafana
Datadog
CloudWatch
ELK
Sentry
Languages & Data
Python
TypeScript
Java
Bash
PostgreSQL
MySQL
MSSQL
MongoDB
Redis
Projects
Selected work
African Stores Canada
Personal
Helping the African diaspora find grocery stores stocking familiar food across Canada.
Production LangGraph agentic pipeline on AWS Bedrock (Claude Haiku 4.5) — typed StateGraph, native tool calling and MongoDB Atlas checkpointing for crash-resumable reasoning. The agent autonomously searches, scrapes and catalogues over 500 businesses across all 10 provinces at under $0.50 per national crawl. Live Next.js web app with a password-gated /ops dashboard and a companion iOS app on the App Store.
Preserving the Igbo language for the next generation — a heritage project for my family.
Rebuilt on Vertex AI RAG Engine (text-embedding-005, Gemini 2.0 Flash), migrated off a local FAISS/Ollama stack and provisioned end-to-end with Terraform + OIDC. Ingested 1M curated Igbo-English pairs from a 19.5M-row NLLB dataset, reaching a RAGAS answer-relevancy of 0.833, served via FastAPI with retrieval scoring and citations.
My wife and I both carry Igbo heritage. This one is for our children.
Cutting through SAST/SCA security-scan noise so engineers only see findings that matter.
A reusable GitHub Actions composite action pipes scan output through a LangGraph agent on AWS Bedrock to classify true positives vs. noise and post structured PR summaries — hardened with OIDC and SHA-pinned actions. The same triage logic is exposed as an MCP server (ci-triage-mcp), so any MCP client can query results interactively (validated via Claude Desktop). The LangGraph state machine keeps every decision auditable and reproducible.
Catching Kubernetes cost and resource-usage anomalies before they turn into waste.
A read-only LangGraph supervisor-worker agent on AWS Bedrock that compares live cluster metrics against peer workloads in the same namespace, then drafts a plain-English explanation and remediation suggestion (rightsizing, unbounded limits, orphaned PVCs) for a human to act on — never auto-applied. Exposed as an MCP server for multi-cluster querying via Claude Desktop. Access is locked to a read-only ServiceAccount (get/list/watch only) with RBAC as code and live-binding verification.
Detecting anomalies in application logs with an automated, secure MLOps pipeline.
End-to-end AWS MLOps pipeline on SageMaker Pipelines (preprocess → train → evaluate → register) deploying an Isolation Forest model with promotion gated on F1 score. Terraform-managed infra with HCP remote state and GitHub OIDC — no static credentials — and multi-stage security gates (Bandit, CodeQL, OWASP, SonarCloud, Trivy) across AKS and AWS.
Turning photos of receipts into structured expense data for family budgets.
Published iOS app integrating Google Gemini for OCR-based receipt scanning and automatic expense extraction. Secure auth with 2FA, email verification and role-based access for multi-tenant family accounts, plus an analytics dashboard with spending insights and export.
