Deep Learning Engineer

Quidient is a deep tech AI company pioneering advancements in Generalized (5D) Scene Reconstruction (GSR). GSR is poised to become one of the world’s great digital product categories (think GPS, MRI, and LMM). Our flagship GSR product, Quidient Reality®, is a powerful API that enables anyone with a mobile device to virtualize, visualize, and measure anything. Words relevant to Quidient include Generative AI, Physics-Informed AI, Large Scene Models (LSMs), Large World Models (LWMs), and API-First.

Overview

We are seeking a Staff Deep Learning Research Engineer to design, build, and train novel neural network architectures that solve hard problems across Quidient’s GSR platform. This is not an applied-ML role — you will work from foundational principles to create new networks from scratch, implement cutting-edge papers, and run end-to-end experiments across domains including geometric anomaly detection, neural rendering, and 3D reconstruction quality.

This is a hybrid position, meaning that you will need to live within easy driving distance to our Technology Center in Columbia, Maryland.

What You’ll Do

Research & Network Design

• Design and train novel deep neural network architectures from scratch for a variety of reconstruction tasks — including surface anomaly detection (e.g., dent detection), geometry-based defect identification, and neural rendering improvements.
• Implement state-of-the-art papers and adapt published architectures to Quidient’s specific reconstruction challenges, exercising deep judgment about what will translate from benchmark to production.
• Identify technical gaps in the current reconstruction pipeline, propose neural network-based solutions, and build the roadmap for how deep learning capabilities evolve across the platform.
• Design and maintain rigorous evaluation pipelines grounded in real-world captures to measure model performance, regression, and generalization.

Model Development

• Run end-to-end experiments independently — from hypothesis through data preparation, training, evaluation, and iteration — with minimal supervision.
• Stay current with the latest advances in deep neural network architectures, training techniques, and optimization methods, continuously bringing relevant ideas into the pipeline.
• Contribute production-quality C++ and Python to integrate trained models into the reconstruction engine.
• Bridge deep learning methods with the geometric and physical foundations of the reconstruction platform, applying domain expertise in one or more of: light transport, 3D reconstruction, or SLAM
• Drive inference optimization and GPU/CUDA performance work toward real-time and on-device targets.