ROIBAN AI

Case studies

The one system I can show you end to end

Twenty years of engagements sit behind this page — most under someone else’s logo and someone else’s NDA. The record is below. The system I can open all the way up is my own: the agent fleet that runs this shop daily.

The record

What the engagements actually built

Figures out of other people’s production are the easiest thing in this trade to inflate and the hardest for you to check. The numbers stay with the clients who own them; what is listed here is the work.

Fig. 02 — delivery diagram20 years of engagements
delivered = scoped is the diagonalmeasured on live systems
Multi‑region
AWS, US-East / US-West
Audited
SOC II visibility + trails
Owned
Email sending reputation
Blue‑green
Deploys with DNS failover

When I say a system will hold, it holds. What I will not do is hand you a figure out of someone else’s production and ask you to take it on faith.

Track record

From the ledger

EngagementDomainMeasureResult
Multi-region AWS migration (PCI)CloudAvailabilityMulti-region cutover · US-East / US-West
Same migrationCloudInfrastructure costConsolidated footprint, lower run-rate
Same migrationComplianceSOC II supportReal-time visibility + audit trails
Digital transformation (The Weiss Group)WebDeliveryCloudFront CDN + asset pipeline
Same transformationEngineeringPipelinesAutomated testing, code review, deploys
Email infrastructure rebuildDeliverabilityDeliverabilitySelf-hosted Mautic · automated IP warming
Pharmaceutical B2B platform (RX Automation)ComplianceControlled-substance ordersFull audit trails (HIPAA-relevant)
NBA streaming app (Roku)MediaPerformanceProduction performance monitoring
Retention analytics platformDataVisibilityLooker / Tableau / Grafana + automated alerts
Deployment architectureOpsRelease safetyBlue-green deploys with DNS failover

Every row is a real engagement. References and fuller write-ups available on request — [email protected].

Selected stories

Three engagements, end to end

Cloud · PCI

The migration that couldn’t blink

As Senior Software Engineer at PCI, I moved production to multi-region AWS across US-East and US-West — a cutover with no downtime budget to spend, real-time visibility and audit trails built in for SOC II, and roughly 40% off the infrastructure bill.

Web · The Weiss Group

The rebuild that made it fast

As Application Development Manager, I led a digital transformation that rebuilt delivery on a CloudFront CDN — and left behind automated testing, code review, and deploy pipelines so it stayed fast.

Email · deliverability

The inbox problem

A deliverability crisis had put most of the list out of reach. Self-hosted Mautic with automated IP warming brought it back — on sending infrastructure the business owns rather than rents.

Live practice

The agent fleet that runs my own shop

Not client work — my own operation, run daily as an agent fleet. It is the standing test bed for everything on the services page.

45+ projects, one fleet

Claude agent sessions across project folders — this site, AWS infrastructure, campaign tooling, quant research — each registered in a session registry, opened and closed by playbook-driven rituals.

One shared memory

A mem0 memory service every agent reads and writes — on the order of 150 durable facts per project namespace — so session 21 starts where session 20 stopped.

Agent-to-agent dispatch

Per-project dispatch channels: agents leave work for each other in mailbox files, and a hub session routes anything that crosses project lines.

Self-verifying writes

Every durable write goes through tooling that rewrites, fsyncs, and re-reads the file byte-for-byte before it counts. Agents that write files need the same guardrails as agents that move money.

Own-ops practice, labeled as such — shown here because it is the one agentic system I can demo end to end without an NDA.

Methodology

How the work gets checked

Instrumented from day one

Monitoring and analytics go in with the first deploy. If it isn’t measured, it didn’t happen — and it can’t be defended later.

Business metrics, not vanity

Availability, latency, cost, deliverability. Quality is monitored through architecture, testing, and analytics — and forecasts get graded on calibration, not on how confident they sounded.

Nothing here I can’t show you

Client figures stay with the clients. The one system on this page I can open end to end — append-only logs, failure-mode catalog, invariant registry — is my own fleet, and it is labeled as own practice, not client work.

Have a system that should exist but doesn’t?

Tell me what you’re trying to do. First conversation is free.