Careers
Engineering for systems nobody else will touch.
The work
NeuraBloq does a specific kind of machine-learning engineering: moving production models and pipelines off Apache MXNet — a retired framework — onto PyTorch and the ONNX ecosystem, and keeping the ones that can't move yet running safely. It is forensic, verification-heavy work: reading Gluon code the original authors left behind, building numerical-parity harnesses, and proving that a ported model is the same model.
The work rewards engineers who like understanding systems more than starting them. If you've ever traced a one-in-a-thousand numerical divergence to a BatchNorm epsilon, or you think "it compiles" is where validation starts rather than ends, this is your kind of problem.
We are a US-based consultancy and work is billed hourly to clients, which keeps incentives clean: we're paid for engineering, not for optimistic estimates. When we have open roles they appear below; if the work described here fits you and no role is listed, a thoughtful note through the contact form is welcome.
Evidence over assertion
Claims about a ported model are backed by a harness and a tolerance, or they are not made.
Respect for running systems
Someone depends on this code today. Stabilize first, migrate deliberately, never rewrite for sport.
Say the hard thing early
Retraining costs, murky provenance, models that should be retired — surfaced when found, not when billed.
Senior, hands-on, US-based
Small teams doing the work directly, with the client engineers rather than over a wall from them.
Join The Team. Apply Now.
Email: info@neurabloq.com