Data infrastructure that catches its own mistakes
I build data platforms for companies that run physical things, and hand them over so they keep working after I’ve gone. What stays behind is software that checks the work: tests, drift alerts, and an agent that diagnoses failures and drafts the fix for a human to approve.
| 842 h | Dev compute added in one week |
| 81% | Scheduled rebuilds colliding |
| 0 | Capacity upgrades needed |
A warehouse’s dev compute jumped 842 hours in a week after an ERP go-live. 81% was scheduled rebuilds colliding. Fix: lanes and schedules. Capacity upgrades needed: zero.
I build the platform, then hand it over.
Most data teams drown in maintenance, and most consultants price by the month and stay. I do neither. The build is fixed price, quoted before anything starts. What stays behind is software that checks the work, and a runbook for whoever runs it next.
A pipeline fails at 3am: the agent in your account diagnoses the root cause, writes the fix, runs the tests, and opens a pull request. The warehouse bill spikes: it finds the queries responsible and proposes the fix. Nothing merges on a model's say-so.
The agent runs in your cloud account, under a role you create. It reads by default. Every change it proposes is a pull request a person approves. No credentials or data leave your account.
↳ A platform that fixes itself, in your account, on your stack.
Pipelines, modelling, and infrastructure on dbt, Snowflake, Airflow, Terraform, or whatever you already run, with lineage and tests from day one.
↳ A warehouse you can hand over, with no one person holding it up.
Semantic layers, dashboards, and metrics that trace to source systems, not to another spreadsheet. Numbers you can put in front of a regulator or a board.
↳ Every figure defensible from dashboard back to source.
Deterministic at the edges, judgment in the middle.
Code watches and code acts; the model only judges in between. Everything on a schedule or a trigger is deterministic code. The model sits in the middle, diagnosing and drafting. Nothing material merges without my sign-off. Speed from the machine, accountability from a person you can name.
Agents diagnose, fix, verify, and open the PR before you wake.
I sign off every material change. Nothing merges itself.
Measured on your systems: pipeline failures fixed before standup, warehouse spend month on month, every check that fired and what it caught.
Built for operators of physical things.
I trained as a mechanical engineer before data, which is why the work starts from the asset, not the dashboard. Electricity distribution is what this method is built for, and where I’m concentrating. The delivered work so far is in mining, automotive and retail.
Operational data verified from SCADA historian to AER submission: availability the historian will back, derating signatures explained, capex justification the regulator will fund.
People call at one of four moments.
The person who catches the mistakes is walking out the door, and what they know never reached the asset register.
A number was wrong and somebody outside saw it first: a customer, an auditor, a regulator.
You are the last check on your own reporting, and it costs you a week every month.
The warehouse spend doubled and nobody can say which queries did it.
If the problem is real, let's talk about it.
I don't do capability presentations. Describe the pipeline that keeps failing, the bill that keeps growing, or the check that lives in one person's head, and I'll tell you whether I can help.
William Marzella, Melbourne — hello@motis.group
I trade as Stygian Consulting, a business name of Tenebrae Holdings Pty Ltd, ABN 44 690 270 645.