Infrastructure · Automation · Applied AI
Infrastructure that builds, heals and explains itself.
Black Mesa designs the whole machine, not just the tool. Cloud platforms as code, operations that run unattended, monitoring that means something, and AI that does real work.
When something says "that's not possible", that's usually where we start.
- Clouds
- AWS · Azure · GCP
- Pipelines
- GitHub · ADO · CodePipeline
- Response
- an engineer, not a queue
Services
Five ways to make the machine run itself
Each one stands alone. Together they close the loop: infrastructure is declared, operations execute unattended, monitoring proves it worked, and AI learns the shape of your estate.
01
Infrastructure as Code
One pull request. An entire landing zone.
Terraform and Terragrunt are the easy part. The package is versioned modules, CI/CD on your provider of choice, and the glue scripts that make it all tick.
02
Automation
Operations that run while you sleep.
Patch nights, server builds, firewall resets and out-of-hours changes that test themselves, roll themselves back, and tell you how it went.
03
Monitoring
Know before your customers do.
Anyone can install a monitoring tool. The difference is what it buys you: fewer 3am calls, evidence for every change, and checks the vendor said were impossible.
04
Applied AI
AI that does work, not demos.
Prompt engineering is the visible ten percent. The rest is retrieval, tools, evals and guardrails that turn a model into a colleague you can trust with production.
05
AI-Built Web
You are looking at one.
Design directions in hours, not weeks. Real code, real performance, and a site that can be reshaped as fast as your business changes.
Why Black Mesa
"That's not possible" is a to‑do list.
Most of what we've built started with a tool, a vendor or a document insisting something couldn't be done. We don't accept that at face value. We read the source, trace the API, and keep going until it works or we can prove why it can't.
A surprising number of the fixes in our history exist because we refused to believe a system worked the way it claimed to. It usually didn't.
You talk to the engineer who does the work. No account managers, no hand-offs, no "let me check with the team".
Open source first
lower cost, no lock-in
Whole package
not just the tool
Knowledge transfer
you keep the keys
- 01
Terraform can't keep module versions in sync across dozens of deployment repos.
✓ Terragrunt tooling that bumps versions itself the moment a module is released.
- 02
Peering config has to be hand-written for every environment.
✓ Generated from one source of truth. Add a network, the peering appears.
- 03
Firewall rules live in the portal, that's just how it is.
✓ Rules in a CSV, validated in CI, applied by the pipeline, reviewed like code.
- 04
Patching needs a change window and a human on the console.
✓ A 12-step unattended runbook, ticket and monitoring blackout included.
- 05
The monitoring product doesn't support that check.
✓ Then we write the check.
- 06
A site like this needs an agency and a quarter.
✓ You're looking at the counter-argument.
How it runs
From 'can we?' to 'it's live' without the theatre
- 01
Listen
day 1
A conversation about what hurts, what's been tried, and what you've been told can't be done. Usually one call.
- 02
Prototype
week 1
Something real in your environment within days, not a slide deck. If the idea doesn't survive contact with reality, better to know now.
- 03
Ship
weeks, not quarters
Built as code, in your repos, on your pipelines. Every change reviewed, every apply logged, every check green before we call it done.
- 04
Hand over
yours to keep
Documentation that writes itself, a walkthrough with your team, and an engineer who still picks up the phone afterwards.
Field notes
Written from the console
The Cloud Has No Undo Button
Azure will back up your VMs, your disks and your databases. It will not back up the network they run on. So I put the whole platform in git — one file per resource, one commit a day.
Peering, Part 3: Overrides, Exclusions, and Importing a Legacy Estate
Generating uniform infrastructure is easy. Generating infrastructure that matches what is already deployed is the job.
Peering, Part 2: The Schema That Made It Generatable
A symmetric object, a boolean called module_output, and a module that deliberately creates nothing.
Next step
Tell us what you were told isn't possible.
A short conversation is enough to know whether we can help. No sales deck, no discovery workshop, just an engineer on the other end.