Forward-Deployed Engineering
Inside the customer's systems.
Accountable through to production.
On these engagements I worked inside the customer's own environment, against live systems and real data, and stayed with the work until it ran in production. This page summarizes how that went and points to the written case study behind every claim.
How I worked
Three habits across every engagement
What stayed constant whether the work was a landing zone, a database migration, or an assessment.
Worked with the operators
I worked alongside the people who ran the systems, which surfaced the legacy behavior, data quirks, and compliance constraints that requirements documents leave out.
Built against real environments
The designs and code were written and tested against the real systems and security rules: database synchronization, failover orchestration, and infrastructure code.
Left evidence behind
Each engagement ended with something the customer's team could run and check: tested infrastructure code, runbooks, and recorded design decisions.
Customer engagements
Work done under real production constraints
Sectors where downtime carries a high financial or public cost. Customer names are withheld; the case studies describe each design and its trade-offs.
Electronic tolling infrastructure
Designed one governed AWS foundation under 40+ accounts, with a hub-and-spoke Transit Gateway network and Aurora disaster-recovery environments kept in parity with the primary.
Telecom platform
Moved a legacy database tier to Aurora with continuous replication and connection pooling, and automated failover while preserving a node-locked software license.
Regulated healthcare platform
Decomposed an on-premise clinical monolith, integrated medical devices over HL7, and shipped through a gated pipeline. The cost work identified $300K+ in annual infrastructure savings across engagements.
AWS migration assessments
Ran VMware discovery from RVTools exports and an agentless collector, bridged an enterprise identity provider to IAM Identity Center, and modeled a Tanzu-to-EKS move with Karpenter. The compute savings are projections, not realized figures.
Incident root-cause analysis
Traced live failures to their low-level causes, from silent call-record ingestion loss to EKS ingress saturation, and wrote them up with the evidence and the fix.
Built end to end
Products I built and run myself
Two products I designed, built, secured, and operate alone, under a written working protocol for AI coding tools. They are not client deployments.
PDF product, hardened and self-hosted
Took a serverless deployment that did not run to a self-audited WebAssembly engine, then moved it off Vercel to a self-hosted setup with verified payment handling.
AI-assisted HR management tool
Leave, attendance, shifts, timesheets, and policy documents in one place. The assistant answers policy questions by quoting the source document and page.
Get in touch
Questions about any
of this work?
I'm happy to talk through any case study: the architecture, the trade-offs, or why I made a particular call.