Senior Platform Engineer · Systems Architecture & AI Infrastructure Research

Franklin Okpako

Building and evaluating infrastructure for reliable autonomous AI systems.

My work connects production systems and distributed architecture to AI inference, trustworthy agent infrastructure and usable business workflows. At BlockOps I build model infrastructure; through Agent Rails I investigate the controls around agents; with businesses I am testing how those components can become useful systems.

The systems stack
01Inference & serving
02Distributed infrastructure
03Authorization · memory · observability
04Usable agent systems
About a decade
in IT, with about 8 years in DevOps and SRE
5 AWS EKS clusters
2 production, platform ownership at Exodus
35+ services · 10+ teams
self-service GitOps platform at Exodus
Millions of users
Exodus wallet and blockchain services, 2022 to 2026
28 s to 60 ms
miniserve median time to first token, sequential vs continuous batching, one run

Built on

AWS
Kubernetes
Linux
Cilium
Terraform
ArgoCD
GitHub
Docker
GitLab
Cloudflare
[ 001 about ]

About the researcher

I began in systems engineering and moved through networking, cloud engineering, DevOps, security, platform engineering and infrastructure architecture. My focus increasingly became designing observable, fault-tolerant, self-healing systems that reduce unnecessary operational intervention and give teams safe, self-service platforms to build on.

At Exodus, from February 2022 to August 2026, I designed and operated Kubernetes and distributed infrastructure across five AWS EKS clusters, two of them production, supporting wallet and blockchain systems at global scale. That work included GitOps architecture, networking, stateful workloads, observability and incident response, and turning recurring failures into automation and more resilient system design.

Today, my applied work at BlockOps focuses on AI infrastructure: model serving, GPU-backed workloads, isolation, observability and simplifying deployment across cloud and privately managed environments. I am going deeper into inference and serving internals so that design decisions follow from how the components actually work.

Alongside that work, I build Agent Rails, an open-source effort exploring authorization, memory, observability, delegated authority, orchestration and evaluation for autonomous AI systems. Agent Guard is the flagship: a concrete place to examine policy, identity, human approval, audit and execution boundaries together.

I am also working with businesses to understand how these components can become usable AI tools. The SME Agent Template supports two different workflow shapes; workflow validation with businesses and development toward an MVP are in progress, not completed real-world validation.

Read the full research vision
[ 002 two research directions ]

Two directions that came out of production work.

Systems for AI

How should AI inference systems schedule compute, memory and requests as workloads scale?

Inference servingScheduling and batchingKV-cache and memoryDistributed AI infrastructure
Evidenceminiserve

Dependable autonomous systems

What has to sit around software that acts on its own before it can be trusted to act?

AuthorizationMemoryObservabilityEvaluation
[ 003 research systems ]

Research systems: questions with running code.

Three systems, each built to answer a question: what batching and KV-cache management change in inference serving (miniserve), what boundary makes an agent's actions safe to authorize (Agent Guard), and how long-lived agents keep memory without it going stale or contradicting itself (MemKit). Agent Rails is the umbrella for the agent-side work, and Vigilo and the rest are in the full list.

[ 004 architecture decisions ]

Architecture decisions: what I chose, and why.

Three decisions from production platforms. Each says what the problem was, what I chose, what it cost where that applies, and what happened. The rest of the experience is on the CV.

Exodus

Self-service GitOps, designed around how teams already worked

Problem
Delivery ran on GitLab push-deploys, and every team had its own habits for building, promoting and rolling back.
Decision
Treat it as a developer self-service initiative, not a tool swap. Evaluate the architecture first, then build reusable workflows for builds, feature environments, promotion and rollback that fit how each team already worked, instead of imposing one ideal process.
Outcome
10+ teams run 35+ services through it without managing the Kubernetes and GitOps underneath, and the migration caused no customer downtime.
ArgoCDArgo RolloutsGitHub ActionsHelmKustomize
Exodus

A fail-closed guard after a reconciliation failure deleted live workloads

Problem
An Argo CD ApplicationSet reconciliation failure deleted live StatefulSets: a generator flap led an orphaned Application's finalizer to cascade-delete the workloads.
Decision
Find the root cause instead of patching one application, and fix it once for the whole fleet.
Outcome
Safeguards now protect workloads across 46+ managed Applications while preserving intended permission revocation.
Argo CDApplicationSetsIncident review
Exodus

A Kubernetes operator so recovery stops depending on one person

Problem
Blockchain nodes serving live wallet traffic failed, and recovery meant someone logging in to repair them by hand. Nobody owned it, and it kept coming up.
Decision
Build a Kubernetes operator that reconciles failed workloads, preserves persistent storage and reschedules with failure-domain awareness, and write short recovery docs for the cases automation cannot safely handle.
Outcome
It replaced part of a recurring manual recovery path and made multi-replica, failure-domain-aware deployments possible.
KubernetesTypeScriptPersistent volumesFailure domains
Full experience on the CV
[ 007 journal ]

Notes from the workbench.

2026-08-01

August 2026: Building memory isn't enough

The harder problem is deciding what should be remembered, and what should be allowed to fade.

2026-07-05

July 2026: Uncertainty is information, not a bug

The most interesting failures are not when an agent is wrong, but when it is uncertain and acts anyway.

[ 008 open source ]

Agent Rails, organized by research problem.

The repositories are components of one research effort. Agent Guard is the flagship; the other components investigate the memory, evidence and coordination it needs around it.

Memory

What should an agent retain, revise or discard when sources disagree?

Observability / evidence

What happened, what was claimed, and what can the record establish?

Orchestration

How do these boundaries hold across a complete workflow?

[ 009 professional work ]

Years of production systems, as evidence for the research.

At Exodus, I operated five EKS clusters and built distributed infrastructure and self-service delivery platforms. At BlockOps, I now apply that experience to model serving and deployment infrastructure. The recurring goal is observable, fault-tolerant systems that reduce unnecessary intervention.

AWS · Kubernetes · Linux · Cilium · Terraform · ArgoCD · Cloudflare · reliability · security

[ 010 get in touch ]

Hiring, research, or a technical question?

The CV has the full record. For roles, collaborations, or consulting on agentic AI infrastructure, write to me.