Joshua Urban Davis — Résumé Download PDF

Joshua Urban Davis

www.joshurbandavis.com | San Francisco, CA 94110 | joshurbandavis@gmail.com

Computer Vision/Machine Learning Engineer & CS PhD (Dartmouth) specializing in the intersection of Real-Time Perception and Human-Centric AI. 6 years of experience building real-time end-to-end CV/ML systems at small startups and big tech (Meta Reality Labs) with a track record of translating high-level research into production-ready systems (3 Patents, Meta Reality Labs, Adobe Research). Pre-PhD, worked at several startups including Bregmann Media Labs, Hex Plexus between 2013 - 2017.

Languages: Python, C++, Swift, JavaScript, C#, MATLAB

Frameworks & Libraries: PyTorch, NumPy, OpenCV, MediaPipe, OpenGL, TensorFlow, scikit-learn, Pandas, ROS2, Reality Kit

Tools: Git, Docker, Unity, Jira, CI/CD pipelines

Specialties: 3D reconstruction, SLAM, lighting estimation, semantic scene understanding, on-device inference, generative AI, AR/VR

APPLECupertino, CA
Applied AI EngineerJun 2026 - Present
  • Build system software, frameworks, and end-user interfaces in C++ to deliver Apple Intelligence features across the stack, from low-level infrastructure to user-facing experiences.
  • Evaluate, iterate, and integrate AI/ML algorithms, including on-device inference, for production deployment across Apple's ecosystem.
  • Collaborate cross-functionally with design, algorithms, privacy, security, performance, and hardware/silicon teams to engineer end-to-end solutions for features used by millions of users.
  • Apply UX intuition to identify experience opportunities, iterate with design, and translate customer needs into technical requirements.
  • Debug and triage complex cross-device interactions under strict privacy and performance constraints.
OVERLAY ROBOTS (pre-seed, stealth startup)Oakland, CA
Founding Computer Vision and Machine Learning Engineer (Lead)Sept 2025 - June 2026
  • Architected and built the end-to-end perception stack as the founding CV/ML engineer for the world's first intelligent makeup robot
  • Drove sensor selection, sensor fusion, and camera calibration for real-time high precision 3D reconstruction, pose estimation, eye tracking, alignment, and skin surface mapping deployed on embedded hardware (Nvidia [Jetson/Orin]) under strict latency and safety constraints.
  • Built data collection, labeling, and evaluation infrastructure from scratch: automated capture pipelines for face scans
  • Defined the perception roadmap and prioritized technical investments across 3D reconstruction, tracking, color perception, and safety monitoring, making architectural decisions that balanced prototype speed with production scalability as the system moved from lab demos to manufacturing-ready hardware.
META | REALITY LABMenlo Park, CA
Research Scientist and Software EngineerMarch 2024 - Sept 2025
  • Designed real-time CV/ML lighting estimation pipeline achieving 6-9ms inference on device, enabling photorealistic virtual object compositing at 72/90 FPS across 25+ indoor/outdoor scene categories in MR passthrough mode.
  • Research, design, and develop applied color perception algorithms for camera and AR applications including novel metric development and key performance indicator evaluation pipelines
  • Developed camera calibration and algorithms for 3D surface reconstruction and semantic scene understanding for multi-sensor perception pipeline, fusing dToF and wide-angle RGB sensors to enable depth completion, sensor calibration, and cross-modal feature alignment
  • Implement and support large video, audio, and depth dataset generation (~15,000 samples) and processing for algorithm development (AI/ML) in new technology explorations.
  • Integrate SLAM, pix2pix, and SIFT algorithms for novel lighting estimation and developed evaluation pipelines for nightly benchmark runs in C++, Python, OpenCV, and internal libraries.
  • Managed full-lifecycle deployment, from model training in PyTorch to high-performance C++ implementation on embedded targets.
ADOBE RESEARCHSan Jose, CA
Research Assistant2021 - 2024 Extended Collaboration
2021 Summer Internship
  • Led the development of media augmented video conferencing creativity tools that leveraged body pose recognition, gesture and speech detection to deliver immersive experiences using Python, Javascript, MediaPipe, and OpenCV.
  • Designed and executed comprehensive mixed-method qualitative and quantitative studies involving 46 users and translated research findings into actionable recommendations.
  • Optimized models for on-device object detection and 3D human understanding running at 60/72 FPS on consumer laptops combining MediaPipe pose estimation, custom gesture classifier (94% accuracy on 18-class vocabulary), and speech-triggered augmentation with <150ms end-to-end latency.
  • Moved work to publication and incorporated into thesis research.
MICROSOFT RESEARCHRedmond, WA
Software Engineering and Research InternSummer 2020
  • Developed and implemented an interactive AR mask to display real-time video of the user's mouth on the mask's surface.
  • Developed perspective-corrected UV mapping algorithm compensating for mask curvature and viewer angle, achieving <3 px distortion error at 72/90 FPS; optimized inference pipeline memory footprint by 40% for sustained on-device operation.
  • Led perception software integration with hardware and accessibility research teams to transition from research prototype to commercial-ready product, addressing thermal constraints, power budget, and sustained inference stability for continuous operation.
  • Led a mixed-methods user study with 18 participants utilizing a collaborative charades-style protocol, demonstrating a 40% increase in communication accuracy and a 35% reduction in task completion time compared to traditional opaque masks.
AUTODESK RESEARCHToronto, ON
User Interface and Graphics Research InternWinter 2019
  • Spearheaded collaboration research across teams to pioneer cutting-edge techniques for authoring 3D objects using generative AI in VR. Leveraged GANs, Transformers, and Autoencoders to develop novel techniques for human-AI collaboration
  • Prototyped and deployed interactive generative adversarial networks (GANs) specifically tailored for 3D design tasks in virtual reality. Employed tech stack including pyTorch, Pandas, Numpy, and Unity to create user-friendly generative AI authoring experiences.
  • Securing a patent for AI virtual reality interaction techniques and system design and published technical paper in peer-reviewed proceedings. (doi.org/10.1145/3450741.3465260)
EDUCATION
DARTMOUTH COLLEGEHanover, NH
Doctorate of Computer Sciences2024
PATENTS