Sr. ML Engineer, Perception

Aureliussystems
San Francisco, CA
On-site

Who this role is best for

Experienced robotics ML engineers with hands-on deployment skills and a background in perception systems will find this role's emphasis on real-world performance and edge computing compelling.

Best fit for

  • Candidates with a history of deploying ML models on physical hardware and optimizing for real-time constraints.
    — “deploy them on embedded compute, and use range results to decide what to improve next”
  • Individuals who have worked across the full ML pipeline with a focus on sensor data and tracking systems.
    — “Worked across the robotics ML pipeline: data, training, evaluation, deployment, and improvement from real-world results”
  • Engineers with a background in EO/IR perception and sensor fusion, ideally with experience in counter-UAS systems.
    — “turning live EO/IR and other sensor data into reliable detection and tracking on a moving, edge-deployed system”

Things to consider

  • Onsite work in San Francisco is mandatory with no remote or hybrid options available.
    — “This role is onsite 5 days a week in San Francisco”
  • U.S. citizenship is required due to the export-controlled nature of the work.
    — “This role requires access to export-controlled information or items and requires U.S. citizenship”

How to stand out

  • Highlight projects where you deployed models on embedded systems and optimized for real-time performance.
    — “Experience measuring and improving performance against real-time latency and compute constraints”
  • Emphasize your ability to work with EO/IR sensor data and integrate it with other modalities.
    — “turning live EO/IR and other sensor data into reliable detection and tracking”
  • Showcase your experience in defining and documenting perception architectures for robotics systems.
    — “Define the perception architecture across detection, tracking, classification, and EO/IR fusion”
  • Demonstrate your ability to build repeatable pipelines for model improvement using field data.
    — “Build a repeatable pipeline for turning range captures, hard negatives, and field failures into training and evaluation improvements”
  • Include examples of collaborative work with hardware and controls engineers in a physical system context.
    — “Work alongside hardware and controls engineers on range days to validate changes on the system”
Pace · Fast PacedCollaboration · HighAutonomy · HighDecision Impact · Team

Derived from job-description analysis by Serendipath's career intelligence engine.

What success looks like

  • turn live EO/IR and other sensor data into reliable detection and tracking
  • improve performance against real-time latency and compute constraints
Typical background
experience in robotics ML pipeline

Skills & requirements

Required

Machine LearningModel DeploymentPythonC++Real-time Latency And Compute Constraints

Preferred

Eo/ir PerceptionRadar-camera FusionSmall Aerial TargetsTrack Management

Stack & domain

PythonC++Machine LearningEmbedded ComputeEo/irRadar-camera FusionSmall Aerial TargetsTrack ManagementJAXPyTorchOn-device Inference OptimizationAutonomous SystemsCounter-uasMachine Vision

About the role

Original posting from Aureliussystems via Ashby

WHO WE ARE

Aurelius is the autonomous directed energy company.

We are a group of engineers, operators, and company patriots building American-made, laser-based counter-UAS systems that neutralize aerial threats.

Archimedes, our autonomous laser turret, combines a high-power laser with radar and machine vision to find, track, and defeat small drones. We design in San Francisco, manufacture in Detroit, and test onsite, on our own cUAS range.

THE ROLE

You’ll own perception for Archimedes: turning live EO/IR and other sensor data into reliable detection and tracking on a moving, edge-deployed system.

This is a hands-on role across the full ML pipeline. You’ll make decisions about data, train and evaluate models, deploy them on embedded compute, and use range results to decide what to improve next. The immediate work is improving the resilience of an existing perception pipeline; as more systems are deployed, field data will create new opportunities to improve performance.

You’ll work directly with robotics software, controls, and hardware engineers and help mentor perception engineers as the team grows.

MUST-HAVES

  • Personally trained, evaluated, and deployed detection, tracking, or segmentation models on a physical system or live sensor stream.
  • Hands-on multi-object tracking and state estimation (Kalman filters, IMM, data association, track management) on real sensor data.
  • Worked across the robotics ML pipeline: data, training, evaluation, deployment, and improvement from real-world results.
  • Strong Python and comfortable reading and writing C++, including the path from trained model to on-device inference
  • Experience measuring and improving performance against real-time latency and compute constraints.
  • U.S. citizenship, as required for this role’s export-controlled work.

FIRST 6 MONTHS

  • Define the perception architecture across detection, tracking, classification, and EO/IR fusion, documented and agreed with controls and robotics software.
  • Build a repeatable pipeline for turning range captures, hard negatives, and field failures into training and evaluation improvements.
  • Establish on-device latency and performance baselines, then improve them.
  • Diagnose and address early field failure modes, including tiny targets, clutter, ego-motion, and track handoff to controls.
  • Work alongside hardware and controls engineers on range days to validate changes on the system.

NICE TO HAVES

  • EO/IR perception, radar-camera fusion, or small aerial targets.
  • Hands-on with current detector families like YOLO and RT-DETR.
  • Data strategies for physical systems, including sim-to-real or fine-tuning on field data.
  • JAX or PyTorch experience with robotics models.
  • On-device inference optimization against a measured latency budget.
  • Distinctive, self-directed projects that solved a specific technical gap.

NOT A FIT IF

  • Your recent work is mainly language models, recommenders, ranking, search, ads, or tabular forecasting, or offline batch work with no on-device deployment
  • Your background is research only, with no deployment on physical hardware
  • You’ve worked only on one narrow part of the pipeline and don’t want to own the broader perception problem
  • You need remote or hybrid work

HOW WE WORK

At Aurelius we work what we call, 5+X. Our core operating hours are 5 days a week, Monday to Friday, 9am to 6pm Pacific Time.

The X is the honest acknowledgment that hard work sometimes takes more. That can be late nights and weekends during sprints and ramp-up periods or conferences, deployments, and events when the mission calls for it.

This role is onsite 5 days a week in San Francisco.

BENEFITS

At Aurelius Systems, we invest in our people. Our comprehensive, competitive benefits package is available at little to no cost to employees and supports health, recovery, and whatever comes next. Explore Our Benefits https://www.aureliussystems.com/careers#:~:text=Explore%20Our%20Benefits.

EXPORT CONTROL NOTICE

This role requires access to export-controlled information or items and requires U.S. citizenship.

NOT A PERFECT FIT?

Email our Head of Talent, alfredo@aureliussystems.us with the most extraordinary thing you've built or done.

Source: Aureliussystems careers (Ashby)

Similar roles

More roles at Aureliussystems