hero

Work at a Portfolio Company

Perception Technical Lead

Sauron

Sauron

IT
San Francisco, CA, USA
Posted on Friday, June 7, 2024

Who We Are

Sauron is the home security company of the future. Homeowners today lack compelling options when it comes to peace of mind against vulnerabilities, and total command and control of their home; there is no definitive, protective brand in the space. Leveraging cutting-edge AI, sensor technology, and nonlethal deterrence, Sauron brings next-generation technology to homeowners to protect their families and property. Incubated by the serial entrepreneur Kevin Hartz and Atomic, Sauron has raised an $18M seed round from leading venture capital firms and angel investors, including 8VC and Flock Safety CEO Garret Langley, to build the new perception system for the home.

The Role | Perception Technical Lead

We are building a perception stack for the home. This system must be able to provide full contextual awareness by reliably identifying threats in all environmental conditions. False positives lead to a terrible customer experience, while false negatives detract from our goal to provide absolute peace of mind. We are looking for a domain expert to lead design, development, and evaluation for perception, marrying classical approaches and machine learning to drive optimal performance. This role will also be highly collaborative with our hardware team, as we develop sensing requirements to guide off-the-shelf component selection and in-house hardware iteration.


What You Will Do

  • Extract the maximum value from our sensors, fusing all observations available while being robust to occlusions, poor lighting and disguises

  • Lead the collection, labeling, and management of datasets to train and evaluate ML models.

  • Leverage state of the art models for 3D object detection, tracking, facial recognition and semantic scene understanding, and push them to the limits of their performance in this problem domain.

  • Perform trade studies to understand the value of new sensors or processing approaches to system reliability.

  • Analyze the performance of systems both in simulation and using data from deployments in the field to find headroom and devise solutions to reduce it.

  • Probe the inner workings of neural networks to uncover and mitigate edge case failures

  • Design systems that can understand and communicate when they are not working well

  • Contribute to machine learning infrastructure (e.g. distributed training, continuous model integration, data management, and evaluation of production systems)

Who You Are

  • You have 5+ years of professional experience with machine learning for hardware products in a safety-critical field, e.g. aerospace, robotics, medical devices, autonomous vehicles.

  • You are passionate about ML, both robust engineering and research challenges.

  • You have a deep understanding of the theory and practice of modern machine learning techniques.

  • You have a clear grasp of basic linear algebra, optimization, statistics, and algorithms.

  • You are experienced at all facets of training and using deep-learning models, including writing custom layers/operations, optimizing networks for inference on edge compute, reproducibility and evaluation.

  • You have extensive experience working with Pytorch, Tensorflow or other modern deep learning frameworks.

  • You are familiar with the use of VLMs and other multi-modal models for semantic scene understanding and description.

  • You are able to solve complex problems with little supervision.

  • You are an excellent communicator, both written and verbal.

  • You have a generalist mindset and can dive in wherever the bottlenecks are, whether that be spooling up cloud compute services to optimizing for embedded systems.

Nice to have:

  • Experience building high-performance software systems using compiled languages (C/C++/Rust/etc.)

  • Experience with Middleware frameworks such as ROS

  • Experience with build systems such as Bazel, CMake

  • Experience in GPU architecture and CUDA programming