- Building Cursor, Anysphere (Cursor)12
Luke Melas-Kyriazi joined Anysphere to build Cursor and is most interested in training large language models for coding. Melas-Kyriazi completed a PhD at Oxford in the Visual Geometry Group, advised by Andrea Vedaldi, with research across machine learning, computer vision, and natural language processing. During the PhD, Melas-Kyriazi also worked part-time as a visiting researcher at Meta Research, and is co-first author, with Meta and Oxford affiliations, of IM-3D (ICML 2024), which generates 3D objects by combining video-based multi-view generation with Gaussian-splatting reconstruction.123
At a glance
- Current role
- Building Cursor
- Company
- Cursor
- Sector
- Company
- Work listed
- 1
- Sources cited
- 3
- Profile reviewed
Career
Luke Melas-Kyriazi earned a DPhil at Oxford in the Visual Geometry Group, advised by Andrea Vedaldi. During the PhD, Melas-Kyriazi worked part-time as a visiting researcher at Meta Research, and after defending the thesis joined Anysphere to build Cursor, with a focus on training large language models for coding.1
Research
Melas-Kyriazi's computer-vision work centers on 3D reconstruction and generation. With Christian Rupprecht and Vedaldi (and Iro Laina on RealFusion), Melas-Kyriazi presented RealFusion, which reconstructs a full 360-degree object from a single image, and projection-conditioned point cloud diffusion for single-image 3D reconstruction, both at CVPR 2023; IM-3D at ICML 2024 combined multi-view video generation with Gaussian splatting.13
Earlier work includes Deep Spectral Methods for unsupervised semantic segmentation, an oral at CVPR 2022. In language research, Melas-Kyriazi co-authored a benchmark for learning to translate a new language from a single grammar book (ICLR 2024 spotlight) and the Harvard USPTO Patent Dataset (NeurIPS 2023 Datasets and Benchmarks).12
Open-source work
Melas-Kyriazi publishes PyTorch implementations of EfficientNet and the Vision Transformer, along with official code for RealFusion, projection-conditioned point cloud diffusion and Deep Spectral Methods.2
Work and contributions
IM-3D: Iterative Multiview Diffusion and Reconstruction for High-Quality 3D Generation Published research
Co-first author of the ICML 2024 paper, with Meta and University of Oxford affiliations, on text-to-3D generation using video generators for multi-view images and Gaussian splatting for reconstruction.31
Timeline
- Published a first paper at EMNLP 2019, extended from a course project.1
- Presented Deep Spectral Methods for unsupervised segmentation as an oral at CVPR 2022.12
- Presented RealFusion and projection-conditioned point cloud diffusion at CVPR 2023.1
- Released IM-3D, later accepted to ICML 2024, as first author.31
Other documented roles
Roles the sources document that are not confirmed as current.
- Visiting Researcher (part-time, during PhD), Meta Research13
Where they work
- Cursor 4 people in the wiki
Related people
- Sasha RushAlso at Cursor
- Alexander WettigAlso at Cursor
- Lauren TanAlso at Cursor
- Christoph LassnerComputer vision · Image, video, and 3D generation
- Tero KarrasImage, video, and 3D generation · Computer vision
- Tsung-Yi LinComputer vision · Image, video, and 3D generation
- Amir HertzImage, video, and 3D generation · Computer vision
- Jonathan HoImage, video, and 3D generation · Computer vision
- Stefano SoattoLanguage models · Computer vision
- Bilawal SidhuImage, video, and 3D generation · Computer vision
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
3 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026