Careers
Xenqube is a small delivery studio: one architect and two to four senior engineers per client pod, working on AI systems for regulated operators. Three roles are open. The hiring bar and interview steps are below so you can decide before you apply.
Hiring bar
Four things, applied to every role. If two of them are not true yet, apply in a year rather than now.
A system in production with users you did not control, that you were paged for, and that survived a dependency or model change. We ask for it by name in the first call.
Our work starts by watching operators work a queue. Engineers who cannot sit through that, or who design before the eval set exists, do not do well here.
Runbooks, ADRs, and eval reports are deliverables, not chores. Clients receive them. Interviewers read a design doc you have written.
Pods are one architect plus two to four seniors. There is no junior tier to delegate to and no separate support team after go-live.
Interview process
Compensation band is stated in the first call. The take-home is paid. No whiteboard algorithms.
With the Solutions Architect. Walk through one production system you built: architecture, what broke, what you changed. We tell you what we are hiring for and the compensation band.
A scoped problem from real client work with the identifiers removed. Example: design the eval set and shadow-mode scoring for an approval queue. Paid at a fixed rate whether or not we proceed.
You present the take-home to two engineers who will work with you. We push on trade-offs, failure modes, and what you would cut under a deadline.
Ways of working, client exposure, and the first 90 days. Offer within three working days of this step. Reference calls happen after the offer, with your consent.
Open roles
All senior. All full-time. Apply with a link to something you shipped, not a cover letter.
Build agent runtimes, tool integrations, and retrieval pipelines for client workflows in logistics, finance, and media. You own the eval set and the shadow-mode scoring for the systems you ship. Python, Temporal or equivalent orchestration, pgvector or Qdrant, and one production LLM system you can talk about in detail.
Build the approval inbox, audit views, and operator dashboards that ship inside Xenith Sort, Twin, Yard, and Desk. Next.js and TypeScript on the front, FastAPI and Postgres behind it. You will pair with operators during on-site weeks.
Deploy and run private LLM stacks on client hardware and VPCs: vLLM, Kubernetes, GPU scheduling, observability, and the upgrade path when a model or driver changes. You write the runbooks the client's ops team inherits.