← Capabilities

Data engineering

Getting clean data off systems that never meant to give it to you.

Overview

Pipelines that hold up at volume and under time pressure: petabyte-scale log ingestion, vehicle telemetry captured off the CAN bus, and real-time dashboards over data that is still arriving. Instrumentation first — a model can only reason over data that was captured properly.

In practice

  • High-volume ingestion and near-real-time processing
  • Telemetry capture from hardware and vehicles
  • Observability with Grafana, Prometheus and the ELK stack

Where this shows up

  • The telemetry pipeline behind Apex Insights: capture off the car, processed and returned to the pitwall while the session is still running. Ours, and the latency budget is ours to solve.
  • Vehicle telemetry for the ARC4, integrated end to end and feeding real-time AI coaching at a race school.
  • Observability across five live properties — Grafana and Prometheus, with analyzer capacity auto-scaling on live session load.

Behind it: the founder’s background

Oz’s own career rather than the company’s work — running since 1997 and still going. It is here because it is why we can do the above.

  • Petabyte-scale log analysis with detections running in near real time. Volume is what makes this hard: at that scale an inefficient pipeline is not slow, it is impossible.
  • The log pipeline underneath it — Windows Event Forwarding, syslog collection, and network and firewall telemetry.
  • A real-time carbon footprint dashboard covering an entire Fortune 500 games company and its players across every platform — hundreds of millions of sessions resolved into a current number rather than a quarterly report. It won the company’s internal innovation award for project impact in 2022.
  • CAN bus data capture from road and race cars during audio recording sessions for Need for Speed — instrumenting vehicles for data in 2012, which is the same job as the ARC4 telemetry stack today.

Let’s talk

If the data exists but you can’t use it — too much of it, arriving too fast, or never captured cleanly in the first place — that’s the problem we like most. Tell us what you’re trying to see.

Talk through the data