Skip to content
Systra

Work

Look at the work, not the description of it

The team’s enterprise delivery record predates the practice, and that record is what we would expect you to examine. Where the work is public, it is linked. Where it is not, it is described in the terms we would use with the client.

Work you can check

Delivery record

What the team has delivered

Delivered inside the organisations named by sector, before Systra. The wider team record covers:

  • A governed Databricks platform and reporting layer for a US construction contractor
  • Production data services and reusable migration frameworks for a UK insurance group
  • API gateway, payment, messaging, CRM, ERP and document integration in Philippine consumer finance
  • Private Kubernetes, delivery automation, platform security and audit evidence in insurance
  • Azure and AWS operations in aviation, healthcare and medical devices, entertainment, automotive and financial services
  • Production AI applications, private-document retrieval and workflow automation

Aaron and Ralph worked for roughly two years on the same insurer platform from opposite sides. Aaron built data pipelines and API services. Ralph built and operated the Kubernetes platform and the deployment automation that released them. Both worked with API management and mainframe integration. They already know how their application, data and platform decisions affect each other.

Three of them, as the situation was, what was built, and what changed — with no figures the record does not support.

  1. Fleet operator · data platform

    A fleet operator's reports disagreed with each other, because each one was built from an unmodelled source by hand.

    What was built

    Oceanic Atlas: a dimensional model derived from that source, and a metric layer regression-tested against production data. Delivered by a single engineer.

    What changed

    One set of numbers. Every report agrees, and the tests prove it.

    Before
    1. Raw source
    2. Hand-built queries
    3. Reports that disagree
    After
    1. Raw source
    2. Dimensional model
    3. Tested metric layer
    4. Reports that agree

  2. Life insurer · platform and DevSecOps

    A life insurer shipped through manual deployments, with audit evidence assembled after the fact and downtime windows for every migration.

    What was built

    A private Azure Kubernetes platform provisioned through Terraform, standardised Helm deployments, and pipelines with security scanning in the release path, critical findings blocked, and control coverage evidenced to audit.

    What changed

    Zero-downtime releases, with the audit evidence produced by the pipeline itself. Cloud cost came down alongside.

    Before
    1. Manual deploys
    2. Evidence after the fact
    3. Downtime windows
    After
    1. Infrastructure as code
    2. Scanning gates in CI/CD
    3. Zero-downtime release

  3. Construction business · applied AI

    A construction business answered partner questions by hand, from documents each partner was not allowed to see all of.

    What was built

    A seven-agent production workflow with isolated knowledge stores for the two business partners, grounded in the business's own documents, with an evaluation loop that measures whether the answers are right.

    What changed

    Answers cite the source. Each partner sees only their own. Accuracy is measured before and after every release, not assumed.

    Before
    1. Question by email
    2. Someone searches the docs
    3. Answer varies by who replied
    After
    1. Scoped retrieval
    2. Grounded assistant
    3. Evaluation loop

Recognise any of this?

Start a conversation