CARTER·HOME

Database platform engineer · SQL Server · Minneapolis

Jim Carter
builds the floor
things run on.

Since 2014 inside a Fortune 100 retailer’s SQL Server estate — first running the servers, now engineering the platform they run on, on-prem and in AWS. Corp-to-corp the whole time.

Isometric line drawing of a data-centre raised floor with one tile lifted, showing copper cable bundles in the plenum beneath two server racks.
Fig. 1 · Raised floor, plenum openNot to scale
Practice
Carter Consulting, LLC
Client
Fortune 100 retail, since 2014
Focus
SQL Server on AWS
Based
Minneapolis · Remote

Practice

Three things I get called for.

The common thread is databases that can't go down and companies that need them somewhere else by a date that's already been promised.

Line drawing of a gantry crane lifting a server rack from one concrete pad to an identical empty pad.
01 — Migration

SQL Server into AWS, without the rewrite

Lift enterprise SQL Server estates onto EC2 with licensing, HA and performance parity intact. Cutover plans that assume the maintenance window is shorter than anyone wants it to be.

Line drawing of two identical storage tanks joined by copper pipes, with a valve and a pressure gauge between them.
02 — Platform

Running it after the migration

Availability groups, backup and restore strategy that's actually been restore-tested, patching, capacity, and the unglamorous instrumentation that turns 3 a.m. pages into Tuesday tickets.

Line drawing of a robotic arm placing identical server blades into a chassis, one after another.
03 — Automation

PowerShell, and now agents

Build-out and cutover automated end to end in PowerShell and Terraform, because doing it by hand sixty times is how estates drift. In the lab, I’m wrapping MCP servers and LLM tooling around the same operational surface.

Track

SQL Server since 1997. One client since 2014.

Staying put this long is unusual for a contractor. It's also why the architecture decisions land — I've had to live with every one of them.

1997 →

SQL developer and DBA

SQL Server since 6.5, on contract across telecom, energy, healthcare, government and finance. SSIS, warehouse loads, reporting, and data-center migrations.

2014 →

Senior SQL Server DBA

At a Fortune 100 retailer: the databases behind the returns operation, then the enterprise platform. Moved a service-tracking system from a SQL Server 2005 cluster to 2017 Availability Groups, and migrated the first on-prem systems to AWS and Azure.

2023 →

Database Platform Engineer (L3)

Same client, supporting several hundred SQL Server instances and thousands of vCPUs across on-prem and AWS. Availability Group and Distributed AG engineering, migration and decommission automation in PowerShell and Terraform. After the team split in July 2023, became the team’s go-to SQL Server DBA for decisions and delivery; formally named its lead SQL Server DBA in August 2025.

Selected work

What the last few years actually looked like.

Reversing an RDS decision

The program was committed to RDS. I built the cost case for EC2 with bring-your-own-license enterprise licensing, took it to leadership, and the direction changed.

  • Cost modelling over a multi-year horizon
  • BYOL vs. license-included analysis
  • Decision reversed at the leadership level

Backup failures, 25% to 1.6%

SQL Server snapshot backups on AWS were failing across accounts. I drove the remediation with the AWS and infrastructure teams: the production failure rate fell from 25% to 1.6%, non-production to 3%.

  • VSS snapshot backups on EC2
  • Cross-team with AWS and infrastructure
  • Production 25% → 1.6%

A migrated procedure, faster than on-prem

Performance triage for a point-of-sale migration to AWS. I validated the database configuration and drove instance right-sizing; the key stored procedure's p95 went from 1.95 s to 28.3 ms, faster than the 71.7 ms on-prem baseline.

  • Configuration validation and right-sizing
  • p95 1.95 s → 28.3 ms
  • Beat the on-prem baseline

Migration, automated

SQL Server estate moved to EC2 with the build, configuration and cutover scripted rather than hand-run — so the last instance looks exactly like the first.

  • PowerShell and Terraform build and cutover
  • Repeatable, auditable instance standard
  • Drift closed off by construction

Operations, with agents attached

In my own lab: custom MCP servers over infrastructure, monitoring and network telemetry — so incident questions get answered by asking, not by opening nine dashboards.

  • MCP servers over live ops surfaces
  • Infrastructure, network and monitoring tools
  • Built to keep the skills current

Finishing ETL work already in flight

At a Minneapolis investment management firm, I picked up SSIS work the existing development team had been on for a while, and finished it.

  • SSIS work picked up mid-stream
  • Completed what the team had in progress
  • The kind of rescue work I like best

The lab

carter-home.com is a real environment.

The domain isn't decoration. Failover, backup strategy, reverse proxying, identity, 10-gig optics and agent tooling all got tried here before they went anywhere that matters. It's cheaper to be wrong at home.

Exploded-view line drawing of a small ten-inch rack in two bolt-together modules, a network switch and two mini PCs sliding in, copper bolts floating out along dashed lines.
Fig. 2 · 10-inch rack, explodedPETG · M6
Standing environment — summary
LayerWhat's runningNotesState
Compute Proxmox VE on four standalone nodes Ryzen / Intel micro hosts plus an i9 tower Online
GPU node i9 tower, RTX 2080 Ti passed through to a VM Built for local speech models; spare CPU works the video re-encode queue Online
Failover Paired DNS filters, resolvers, reverse proxies and alerting Floating VRRP addresses, each pair split across nodes, failover drilled Online
Storage NAS + Proxmox Backup Server Nightly backups, offsite to Backblaze B2 and a second site Online
Patching Rolling weekly updates, one node at a time Health-gated reboots while the pairs carry traffic Online
Network UniFi gateway, managed 2.5G switching 3 Gbps fiber WAN, cold-spare router on the shelf Online
Backbone Single-mode fiber run between floors 10G BiDi optics, 40 m invisible-fiber pull Online
Edge Traefik pair, OIDC proxy, ACME wildcard certs Nothing internal published unauthenticated Online
Monitoring Netdata, Loki and scheduled health sweeps Known faults get a scripted fix, not a second page Online
Automation Home Assistant + custom MCP servers Infra, network and telemetry tooling Online
Rack 10-inch rack, printed in-house 5U + 3U, PETG, M6 bolt-together In build

Off the clock

Mostly the same instinct, different materials.

Line drawing of a CoreXY 3D printer with a copper honeycomb part half-printed on the bed.
Print fleet

Two FDM printers, open firmware only

A CoreXY on Klipper and a multicolour machine. Open firmware is a hard requirement — a printer I can't read the source of is a printer I can't fix.

Cutaway line drawing of a twin-turbo V8 and dual-clutch gearbox on an engine stand, a copper diagnostic cable running to a laptop.
Garage

Audi diagnostics, done properly

A tuned S6 and an SQ5, both diagnosed on VCDS rather than guessed at. Module addressing, adaptation channels, factory documentation. Same method as a production incident, smaller blast radius.

Line drawing of a repair bench: a circuit board in a helping-hands clamp under a magnifier, a soldering iron's copper tip on one joint.
Bench

Board-level repair

If it's a solder joint, it gets a solder joint. Replacing a working machine over one failed component is how landfills get filled.

Contact

Get in touch.

SQL Server on AWS, something on this page, or something in the lab.