Backend Engineer, ML systems

You'll help build the backend and ML platforms that turn multi-sensor satellite data into geospatial intelligence — and you'll build them to run fully air-gapped and offline , deployed inside defense and intelligence environments with no internet access.

This is an early-career role.

We're not expecting you to have done all of this already.

We're hiring for how fast you learn, how you debug when things are murky, and whether you have the instincts to become genuinely good at a rare intersection: production backend, applied ML, and hard-constraint systems that have to work without the cloud crutches most engineers lean on.

What makes this different

Most backend/ML jobs let you reach for a managed service when things get hard — a hosted model API, cloud autoscaling, pip install at deploy time.

Here you can't.

Systems run on isolated, on-prem hardware with no internet at runtime.

That means:

• Models are self-hosted and run locally — no external inference endpoints.

• Dependencies are mirrored and builds are reproducible and offline-friendly.

• Deployment and updates happen through controlled, secure processes, not push-to-cloud.

• Monitoring, logging, and evals all have to be self-contained.

If that sounds like a fun constraint rather than an annoyance, you'll fit well here.

What you'll actually do

• Build and maintain backend services and APIs that fuse satellite data (SAR + MSI) and serve geospatial analytics to analysts.

• Design and work with databases and data flows for large raster/imagery datasets — model the data, write queries and pipelines that hold up at scale.

• Put ML models behind reliable, self-hosted services: take something that works in a notebook and make it a monitored, production service that runs offline.

• Work on inference and ML pipelines for imagery/geospatial analytics — batching, latency vs. throughput, GPU constraints, keeping things healthy in an air-gapped deployment.

• Contribute to analyst-facing and agentic tooling — components that chain steps, call local tools, and support intelligence workflows (and fail gracefully when they don't).

• Debug systems where "correct" is fuzzy, and build the evals and monitoring that tell us whether a change actually made things better.

What you'll learn here

Because this matters as much as the work:

• How to build production ML and data systems under real constraints — offline, on-prem, security-first — a skill very few engineers ever develop.

• How to reason about systems that are probabilistic-ally correct, not just pass/fail.

• Applied geospatial/EO machine learning and multi-sensor data fusion, mentored by [a lead with ML engineering experience / the team].

• How to stay sharp without managed services — reading source code and papers, and self-hosting what others just call an API for.