Dmitry (Dima) Petrov
🪁 Hi, I'm Dima — a spatial AI researcher and engineer.
I currently work at Foundation EGI, where we automate labor-intensive CAD tasks such as technical drawings and process planning.
I recently defended my PhD (thesis presentation) at the College of Information and Computer Sciences, University of Massachusetts Amherst, where I worked with Evangelos Kalogerakis at the intersection of 3D vision, graphics and AI. My research focused on generative 3D AI — methods that reconstruct and generate 3D shapes with complex geometry and topology while giving fine-grained, interpretable control over their structure and form. Earlier, I worked at the intersection of machine learning and computational neuroscience, on automated quality control of brain segmentation and the analysis of structural connectomes.
Before my PhD I earned a Master's in Data Science with Leonid Zhukov and a Bachelor's in Mathematics with Alexander Kolesnikov, both at the Higher School of Economics (HSE).
Publications
Vision, Graphics & AI
Neuroscience
Note: I no longer actively work in this area — my research has since moved to vision, graphics and AI. These works are listed for completeness.
Experience
- 2025–present Research Engineer, Foundation EGI — on the technical-drawings team, building spatial-AI systems that automate labor-intensive CAD work; work on machining-feature detection in low-data regimes and CAD tokenization.
- 2024 Research Intern, Roblox, San Mateo, CA — developed kinematic-graph inference for segmented 3D meshes, recovering joint connections and parameters so static meshes can be rigged and articulated automatically; resulted in an ICCV publication (GEOPARD) and a patent application.
- 2022 Research Intern, Adobe Research (BIG Lab), remote — designed a generative diffusion model for 3D shapes built on a novel skeleton-based neural field, producing shapes with robust geometry and topology (published as GEM3D, SIGGRAPH 2024).
- 2021 Research Intern, Adobe Research (Procedural Imaging Group), remote — developed part-aware, editable 3D shape reconstruction by assembling local implicit functions, enabling fine-grained edits of reconstructed shapes (published as ANISE, TVCG 2023).
- 2019–2025 Graduate Research & Teaching Assistant, UMass Amherst (CICS), Amherst, MA — conducted PhD research on generative 3D AI (skeleton-based shape generation, part-aware reconstruction, and cross-shape attention for shape analysis); assisted Machine Learning, Visual Computing, and Game Programming classes.
- 2015–2018 Earlier ML research (computational neuroscience) — Imaging Genetics Center, USC: large-scale quality control of brain-shape segmentations and evaluation of 35 structural-connectome construction pipelines; IITP, Moscow: kernel-based neurodegenerative-disease classification from connectomes.
- 2016 Freelance Data Scientist, Ezhome (gardening-services start-up) — built pricing models from web-scraped real-estate data, expanding instant-quote coverage from 50% to 80% of users and lifting conversion ~5%.
Service
Mentoring
- 2020–2024 UGRAD Research Volunteers, UMass CICS, Amherst, MA — mentored undergraduate students on machine/deep learning and data science topics. A recurring CICS program; I took part six times.
- 2020–2021 PhD Applicant Support, UMass Amherst, Amherst, MA — mentored prospective international PhD applicants as part of the PhD applicant support program.
- 2019 Eureka! & Women in Engineering volunteer, Amherst, MA — led "Creative Computing with Scratch" and "LED Programming" workshops for high-school girls from Holyoke as part of Eureka!, a program addressing the gender gap in STEM, and at the Women in Engineering Career Day.
Reviewing
Conferences: CVPR (2023–2026), ICCV (2023–2025), SIGGRAPH (2025–2026), NeurIPS (2024–2026), ICML (2025–2026), ICLR (2025), ECCV (2024, Outstanding Reviewer), WACV (2024–2025), Eurographics (2025).
Journals: TPAMI (2022–2023), TVCG (2022–2025).
Past life (journalism & copywriting)
Before tech, I spent almost a decade in journalism and copywriting. I started in journalism before university — as an FMCG reporter at the business weekly SmartMoney, then a reporter and later editor at the daily newspaper Trud. Later, alongside my undergrad, I freelanced as a copywriter and editor in the advertising and custom-publishing departments of Harvard Business Review Russia and Sanoma Independent Media. I genuinely loved that work. (Fun fact: Trud — "Labor" — once held the Guinness record for the world's highest-circulation newspaper, peaking at over 21 million copies in 1990.) As the space for independent journalism in Russia kept shrinking, I moved to math and computer science — probably the right call, since most of the outlets I wrote for are now gone or state-dependent, though I still miss it sometimes. Earlier still, I dabbled in online marketing for software that ran on Pocket PCs and pre-iPhone smartphones — because why not.
- 2010–2014 Freelance Copywriter, Harvard Business Review Russia (advertising) — advertorials and business-case tasks for magazine competitions.
- 2009–2011 Freelance Editor, Sanoma Independent Media, Moscow (custom publishing) — corporate print issues; won the 2010 PROBA-IPRA "Best Corporate Media" award for PwCCampus.
- 2007–2009 Reporter → Editor, daily newspaper Trud — covered the Russian labor market, employers and occupations; promoted to edit the Labor Market section.
- 2006–2007 Reporter, business weekly SmartMoney — feature articles on FMCG markets and companies; Best Article of 2006 for "Red Directors".
- 2005–2006 Product Manager, Paragon Software — product copy, email campaigns and Google AdWords advertising for the Pocket PC platform.
Some of my old material (in Russian): sostav.ru.