A two-person studio for embedded product iteration, with GCP DevOps wired in from day one. Vertex AI, Gemini Enterprise, multi-model orchestration: built into your codebase, not handed over a fence.
"We don't ship code over the fence."
We work inside your repo, on your branch protections, in your standup. Pull requests on day three; design docs that match what's actually deployed. The engagement ends when your team owns the result.
"From day one. Not retrofitted."
Terraform from first commit. Cloud Build pipelines, Cloud Run services, Pub/Sub fan-out, BigQuery sinks. We don't bolt CI onto a six-month-old repo. We set it up before the first feature lands.
"Multi-model. Vertex AI. Gemini Enterprise agents."
Deterministic routing across Gemini 2.5, Claude Opus 4, and GPT-5, with budget caps in milliseconds and dollars. Vertex AI Agent Engine deployed on your VPC, wired to existing IAM. Eval harness running on day one.
By the time a growing SaaS notices that deploys are slow, observability is patchy, and rollbacks are scary, the cost of fixing it has compounded for two years. We write the Terraform first, the feature second.
We're specialists. Cloud Run, BigQuery, Vertex AI, Pub/Sub, IAM, VPC-SC. These are the surfaces we know cold. Pretending to be cloud-agnostic would mean shipping mediocre work on three clouds instead of excellent work on one.
Models get deprecated. Pricing changes. A new release outperforms last quarter's leader. A routing layer with deterministic fallback isn't optional infrastructure. It's the only way an AI feature stays in production for two years.
Every project ends with a runbook, a Terraform module under your VCS, and zero SaaS 971 logins remaining in your GCP project. A consultant who's load-bearing after they leave is a bug, not a feature.
The biggest shift wasn't Kubernetes, it was visibility. For the first time we can see deployments, system health, and rollbacks in one place, and act on it without outside help.
Eleven years embedded inside one of the US's largest privately held payment acquirers, first as CTO and then as CIO, building production infrastructure on GCP long before managed AI was a slide deck talking point. Vertex AI, AlloyDB, BigQuery, multi-model orchestration: not retrofitted, built in from commit one.
Before payments, Derek founded a UCaaS and cloud application hosting company in the years when "the cloud" still needed scare quotes. Multi-tenant voice, video, and hosted applications for enterprise customers, before AWS had a single-digit service count. It was acquired. The lesson, that infrastructure decisions made on day one follow you for a decade, stuck.
SaaS 971 exists for the gap between demo and production. Most AI engineering work stalls there. This studio doesn't hand things over the fence: it embeds, ships, and leaves teams owning the result.