[I] ML Research Intern - Reconstruction (Spring 2027)

WHO WE ARE:

Peripheral is developing spatial intelligence, starting in live sports and entertainment. Our models generate spatial data, used for advanced sports analytics and immersive media experiences. We’re solving key research challenges in 3D computer vision, creating the foundations for the next generation of robotic perception and embodied intelligence.

We’re backed by top investors, including Khosla Ventures, Inovia, Deloitte Ventures, Daybreak, and Entrepreneurs First, and working with some of the biggest names in sports. Our team includes engineers and researchers from leading technology companies and research institutions, and we’re building technology at the intersection of AI, graphics, and the future of live entertainment. We’re ambitious and looking to win.

THE OPPORTUNITY:

We’re seeking a Research Intern to join Peripheral’s Reconstruction team for a research-focused internship, working on the problems at the heart of how we turn raw multi-sensor capture into rich 3D understanding of the world. You’ll work on a defined research project in 3D reconstruction, feed-forward reconstruction, neural rendering, generative models for spatial data, or world models, with the goal of results that can feed directly into Peripheral’s reconstruction work.

You’ll work closely with our reconstruction team, getting direct exposure to how cutting-edge research is evaluated, pushed forward, and moved toward production at an early-stage company.

WHO YOU ARE:

You’re currently a Master’s, or PhD student in Computer Science, Robotics, Computer Vision, or a related field, with prior research in 3D reconstruction, diffusion models, or world models and the publication record to show for it.

You’re comfortable going from a new idea or paper to a working implementation and an honest evaluation of whether it holds up. You have your own opinions about where the field is headed, and you can back them up with experiments.

You work independently and take ownership of your project, while sharing results openly, welcoming feedback, and learning quickly from the people around you.

You’re excited by the chance to do research that has a real path to being used, and to contribute to Peripheral’s research culture while it’s still being shaped.

WHAT YOU’LL BE DOING:

• Own a research project in 3D reconstruction, feed-forward reconstruction, neural rendering, world models, or diffusion-based modeling for spatial data, from initial idea through experiments and evaluation.

• Implement and evaluate new techniques from the research literature and from your own ideas, and report results clearly to the team.

• Work with real captured data, learning how its characteristics and limitations affect the methods you develop.

• Build the experiment tooling and evaluation infrastructure you need to move quickly and rigorously.

• Present your findings to the team and document your work so it can be built on after your internship.

REQUIREMENTS:

• Currently enrolled in a Bachelor’s, Master’s, or PhD program in Computer Science, Engineering, Robotics, Computer Vision, or a related field.

• Prior publication(s) at a peer-reviewed venue (e.g., CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, SIGGRAPH, 3DV) in 3D reconstruction, diffusion models, or world models.

• Strong research experience in 3D reconstruction, diffusion models, or closely related areas (e.g., feed-forward reconstruction, visual geometry transformers, Gaussian splatting, neural rendering, multi-view geometry, video generation, world models, generative models for 3D).

• Proficiency in Python and a deep learning framework, with a track record of implementing and training models from scratch.

• Strong communication skills, able to explain results and tradeoffs clearly to both researchers and engineers.

• Candidates must have the legal right to work in Canada for the duration of the internship and be based in or willing to relocate to Toronto for an in-office role. At this time, we are unable to provide immigration sponsorship.

NICE TO HAVE:

• Publications at top-tier venues as a first author or major contributor.

• Experience with both classical (multi-view geometry, structure-from-motion) and neural (implicit representations, Gaussian splatting) reconstruction approaches.

• Experience with feed-forward reconstruction models (e.g., VGGT, DUSt3R, MASt3R), monocular depth estimation, or multi-view stereo.

• Experience with world models or video generation models for 3D and spatial understanding.

• Experience with large-scale training, including distributed or multi-GPU workflows.

• Released code, datasets, or open-source contributions associated with your research.

WHY YOU'LL LOVE WORKING HERE:

• High ownership of high-impact projects shaping the future of spatial intelligence and 3D media.

• Mentorship from world-class engineers and researchers.

• Unparalleled access to premier global sporting events and iconic venues.

• Flexible Paid Time Off (PTO).