EMEKA ELUWA / PLATFORM & AGENTIC AI
Open to remote AI-native platform roles

Emeka EluwaPlatform engineer for agentic systems

I build the infrastructure agents run on — and the observability that keeps them honest. Seven years of cloud and platform engineering on Azure, now pointed at LLM systems in production: orchestration, evaluation, tracing, and the failure modes nobody demos.

trace · alert-triage · 3 sourcesstatus 200
flux.notification.eventgitops
datadog.monitor.triggersaas apm
alertmanager.webhookprometheus
normalize.envelopeschema fan-in
context.enrichlogs + host meta
claude.triage.classifyllm
severity.routedeterministic
teams.channel.postteams
Schematic — production alert triage running at CData.
Flux · Datadog · Alertmanager → normalise → enrich → Claude → route → Teams.

span_01 · practice

What I actually do

Most people building agents have never had to keep one running at 3am. I came the other way round — from platform engineering into AI — so the parts that break in production are the parts I think about first.

7 yrsCloud & platform engineering
Azure since 2019
3Claude Code skills authored
and in team use
3Platforms architected
and owned end to end
11+Store integrations running
in production
01 — Platform

Infrastructure that holds

Azure, Kubernetes, Docker, CI/CD. Provisioning, deployment pipelines, and the boring reliability work that decides whether anything else you built matters.

02 — Agents

Systems, not demos

Multi-agent orchestration with LangGraph and CrewAI, RAG over vector stores, MCP tooling, and human-in-the-loop approval gates where autonomy is the wrong default.

03 — AgentOps

Knowing when it's wrong

Tracing, evals, and monitoring for non-deterministic systems. Durable execution, idempotency, and delivery semantics for workflows that run for hours, not milliseconds.

span_02 · systems

Selected systems

Autonomous alert triage pipeline

CData · Platform Engineering In production

An LLM sitting inside the on-call path, fed by three heterogeneous alert producers. Flux, Datadog and Prometheus Alertmanager each emit a different payload shape; the pipeline normalises them into one envelope, enriches it, has Claude classify and summarise, then routes deterministically so only what matters reaches the engineer — with the reasoning attached.

  • Fan-in across three alert sources — FluxCD reconciliation events, Datadog monitors and Prometheus Alertmanager — normalised into a single alert envelope so downstream stages never branch on producer
  • Deterministic routing around a non-deterministic core: the model classifies and summarises; severity and destination are decided by code, so an LLM error cannot misroute or silently drop an alert
  • Measured, not asserted: 333 executions over 11 days — 251 Datadog, 51 Alertmanager, 30 FluxCD — at 49s median enrichment latency and under $0.20 per alert, with an 88% prompt-cache hit rate
  • Built for failure: bounded retries, structured output parsing, and graceful degradation to raw alerts when the model call fails
DatadogPrometheus AlertmanagerFluxCDn8nClaude APIMicrosoft TeamsKubernetesAzure

Authored Claude Code skills

CData · Internal tooling In team use

Three reusable agent capabilities packaged as Claude Code skills — durable tooling that encodes a review standard once and applies it consistently, rather than re-prompting it every time.

  • Security-first code review — threat-model-aware review pass surfacing auth, injection, and secrets-handling issues before they reach a PR
  • General code review — house-standard correctness and maintainability pass
  • Content agent skill — the generation and publishing workflow behind CheckPrice's editorial pipeline
  • Contributing to an internal MCP and Connect AI initiative, extending agent access to live data sources
Claude CodeMCPAgent skillsPrompt architectureAppSec

Rackapress

Founder & principal engineer Pre-launch

A multi-tenant managed hosting control plane for the Nigerian market, operated by Stackoncloud Ltd. Sole architect and implementer — control plane, data plane, provisioning, backups and the operator console.

  • Made cross-tenant access structurally impossible rather than merely forbidden: found mid-build that the incumbent stack ran every tenant under one shared OS user, stopped, moved to container-per-tenant on Incus, and proved non-overlapping UID maps at the kernel level
  • One idempotent worker chain as the only privileged path — every privileged action runs through the same BullMQ chain regardless of trigger, with durable job state that survives restarts and an append-only audit log
  • Numbers, not adjectives, on every gate: real-infrastructure runs caught five defects a green test suite missed, including the shared subuid pool that was the critical isolation bug
  • Capacity proven empirically via an OOM harness on live infrastructure — throttle, hard-kill and zero blast radius measured across all three tiers
  • 3-2-1 backups with restic to offsite storage, under a spec-as-contract workflow: single-concern PRs, no self-merge, and 39 test suites across six CI jobs on every pull request
PostgreSQL 18DrizzleBullMQIncusAnsibleresticNext.jsHetznerPaystack

CheckPrice.ng

Founder & sole engineer Live

An AI-native price intelligence platform for the Nigerian market, aggregating listings across Jumia, Konga, Jiji, Slot and seven other stores. Built mobile-first on lean infrastructure to survive what earlier attempts could not: running cost.

  • Full stack in production — Next.js (App Router, SSR), React Native/Expo, Python/Scrapy scraping infrastructure, Supabase/PostgreSQL, with organic traffic and signups
  • Agent with a human gate: an autonomous content agent that drafts, enriches with imagery, and publishes — behind a propose-then-commit Telegram approval step, because unsupervised publishing is a liability, not a feature
  • Semantic product matching using embeddings to reconcile naming variation across stores
  • Security from day one — row-level security, server-side auth middleware, XSS sanitisation, Cloudflare Turnstile, and managed secrets
Next.jsTypeScriptReact NativePostgreSQLSupabasePython/Scrapyn8nEmbeddingsCloudflare
checkprice.ng ↗

TownRunner

Lead engineer In development

A peer-to-peer delivery marketplace for Lagos's informal economy. Posters, Runners and Receivers coordinate through an escrow and delivery-code trust system — structured as a logistics job board rather than a fleet dispatch service.

  • Three implementation phases shipped: OTP auth, role-based flows, run lifecycle, wallet, real-time GPS over Socket.io, and four Paystack payment variants
  • PostGIS for geospatial queries, with a Redis/BullMQ/PgBouncer-ready architecture that scales from MVP without a redesign
  • Kobo-denomination precision across every payment layer to eliminate floating-point drift
  • Dynamic pricing engine with DB-stored peak and express multipliers for zero-deploy tuning
React NativeNode.jsFastifyPostGISSocket.ioRedisPaystackDocker

span_03 · stack

Stack

Things I've shipped with, not things I've read about.

Cloud & platform

AzureKubernetesDockerAKS (multi-region)TerraformFluxCD GitOpsHelmSealed SecretsEntra ID / Key VaultIncusAnsibleAzure DevOpsHetznerVercelCloudflare

Observability & reliability

DatadogPrometheus / AlertmanagerFluxCD / GitOpsMulti-source alert normalisationStructured loggingTracingIncident responseDurable executionIdempotent job chainsrestic / 3-2-1 backups

Agentic AI

LangGraphLangChainCrewAIAutoGenMCPClaude APIOpenAI APIMulti-agent orchestrationHuman-in-the-loop design

RAG & evaluation

ChromaDBPineconeEmbeddings & semantic searchPrompt engineeringAgent evalsHallucination detection

Languages & backend

PythonTypeScriptJavaScriptSQLNode.jsFastifyNestJSSocket.ioBullMQ

Data & security

PostgreSQLPostGISSupabaseRedis / ValkeyDrizzle ORMBullMQRow-level securityKernel-level tenant isolationSecrets management (libsodium)SOC2-aligned practicesPII redaction

span_04 · history

Where I've done it

May 2025 — Present

Senior Platform Engineer

CData Software · Remote (US-headquartered)

One of eight engineers running the Kubernetes platform behind Connect Cloud, a data-integration SaaS serving 7,000+ enterprise customers across three AKS regions. Terraform, FluxCD GitOps, Helm and Sealed Secrets; production incident response on distroless workloads. Built the LLM alert-enrichment pipeline now in the on-call path and authored three internal Claude Code skills.

2025 — Present

Founder & Principal Engineer

Rackapress (Stackoncloud Ltd) · Lagos, Nigeria

Sole architect of a multi-tenant hosting control plane: schema, provisioning, container isolation, backup strategy and operator console. Every architecture decision written to spec and implemented through single-concern, CI-checked pull requests.

2024 — Present

Founder & Engineer

CheckPrice.ng · Lagos, Nigeria

Designed and built a price intelligence platform from nothing — web, mobile, ingestion infrastructure, and the agentic content pipeline. Sole owner of architecture, security and production stability.

May 2024 — May 2025

Independent Consultant & Technical Mentor

Freelance

Directed and reviewed junior engineers across cloud, Kubernetes and CI/CD — setting approach and holding a review bar rather than producing to spec.

May 2022 — May 2024

Cloud / DevOps Engineer

Plan A Technologies

Owned CI/CD and delivery automation — Docker workflows, PowerShell auto-remediation, Ansible-driven deployment from Artifactory into customer environments. Migrated TeamCity from on-premises to Azure.

Jul 2021 — May 2022

Azure Cloud Engineer

Mczeek LLC

Translated architecture designs into Terraform — Azure databases, firewalls, load balancers — and ran build and release pipelines in Azure DevOps through to deployment.

Jul 2019 — Oct 2021

Azure Cloud Engineer — App Service (L3 / SRE)

Tek Experts (Microsoft support partner)

Resolved escalated L3 cases on the Azure App Service platform for Microsoft's international enterprise customers — debugging production systems I had not built, reasoning from whatever telemetry already existed. Where the observability instincts came from.

Credential

JHU

Certificate Program in Agentic AI

Johns Hopkins University · 2026

18-week program in agent architecture, multi-agent systems, reinforcement learning, RAG, evaluation, and LLMOps. Three hands-on systems built: a RAG decision-intelligence pipeline, a multi-agent research analyst, and an underwriting system in LangGraph.

Education

Undergraduate

BSc, Computer Science

Southwestern University, Nigeria · Dec 2024
Certification

Azure Administrator Associate

Microsoft Certified · AZ-104
Diploma

HND, Electrical & Electronic Engineering

Lagos City Polytechnic · 2015
Emeka Eluwa
Emeka Michael Eluwa
Lagos, Nigeria · UTC+1
Supporting international customers since 2019

span_05 · contact

Let's build something that stays up

Open to remote roles in AI-native platform, AgentOps, and forward-deployed engineering. Also available for consulting on agent infrastructure and observability.