Work · Formats
Embedded systems engineer
Ackren answers in single-digit milliseconds because the latency is architectural, not a tuning result. This role keeps it that way on real hardware, down to sixteen megabytes of flash with no card at all.
The work
What you would do
- 01
Meet the three floors
Format names the hardware floor: Folio on Linux with a memory-mapped, gigabyte-scale artefact, Quarto on PSRAM and SD at roughly two hundred megabytes, Octavo executed in place in sixteen megabytes. Any pack runs on any format, subject only to flash.
- 02
Hold the latency
Keep responses in single-digit milliseconds, measured end to end. Three point six milliseconds on an eighty-word reply is the current figure, and it must not erode under load.
- 03
Live in the budget
Work within fifty kilobytes of SRAM plus a memory-mapped index region. On Octavo the twelve-bit fingerprints give a false accept of one in four thousand and ninety-six, and that weaker guarantee is stated with the name, every time.
- 04
Keep it re-derivable on device
The derivation log has to be producible on the device, not only on a workstation, because the log is the product.
Requirements
What we are looking for
You have shipped code onto memory-constrained targets and you measure before you claim. Comfort with execute-in-place on flash, memory-mapped data structures, PSRAM and SD data paths, and fixed, deterministic arithmetic is what the work needs. You treat a latency number as something you can reproduce on a bench, not something you assert.
- C and embedded
- Memory-constrained targets
- Flash / XIP
- Memory mapping
- Deterministic arithmetic
- Measurement discipline
Terms
How we work
Fully remote, worldwide, with a few hours of overlap with UK hours, GMT or BST. Every figure you publish carries its unit and its basis: measured, computed, modelled or cited. Work you build for Aleten is assigned to Aleten, said here rather than later.
Two things we will not let you overclaim
The three formats are not a capability ladder. Inference cost here is effectively zero, so the logic that justifies a model-size ladder does not exist, and a deck that implies one is wrong. And determinism holds within a build, identified by artefact hash, not across builds. We do not claim bit-identical output everywhere, and neither will you.
Apply
A short form, read by a person
Tell us about a memory-constrained or latency-critical system you have shipped, and how you measured it. Numbers with their basis are the fastest way in.
The form stores your application in our own database and a person reads it; there is no third-party tracking system. If you would rather write, send a line to work@aleten.com saying you want the embedded systems engineer role, and a person will reply.