Oleksandr Maksymets
AI Researcher — Foundation Models, Multimodal Learning, Scalable Training 🇺🇦
Meta Superintelligence Labs (MSL) / Meta AI Research
I’m an AI researcher at Meta Superintelligence Labs (MSL). My work sits at the intersection of foundation models and scalable learning systems—especially multimodal training and vision representation learning.
At FAIR, I helped build AI Habitat and worked on embodied learning problems (navigation, rearrangement, sim-to-real), as well as language grounding in 3D from raw sensor data (Locate3D).
I care about turning large-scale training into reliable capabilities: strong representations, robust generalization, and systematic evaluation—bridging research and the engineering required to make models work in practice.
I organize the CVPR Embodied AI Workshop and review for major ML/CV venues (CVPR/ICCV/ECCV, NeurIPS, ICLR, ICML).
- Foundation Models
- Multimodal Learning
- Representation Learning
- Scalable Training & Evaluation
- Agents / Decision Making
- Embodied AI
-
PhD, Computer Science
Taras Shevchenko National University of Kyiv
-
BSc + MSc (Honors), Applied Mathematics
Taras Shevchenko National University of Kyiv
-
Assistant Professor, Cybernetics Department
Taras Shevchenko National University of Kyiv
Publications
Experience
Natural Language Processing Service for Facebook Search. Working in the following areas:
- Machine learning: training models, feature engineering.
- Distributed systems: building scalable service for producing NLP signals
- Data processing: writing Hive jobs to process large volumes of data.
- Mentoring: helping others to get familiar with systems.
- Trained and launched multilingual phrase/entity embedding with wide impact across the organization.
- Developed number of C++ classes/API for usage of phrase/entity embedding that combine different types of loading and generalize types of embedding.
- Lead and was responsible for an infrastructure/backend of Search NLP Service.
Actively participated in launch of search for post to number of markets in reliable way and succeed to automate scaling search for new locales. Some projects:
- Intensive data-mining, feature engineering, language model training for Search i18n purposes, work with Speller models
- Built tool for text-based feature engineering that advice you most suitable NLP resources (Language models, embedding, etc).
- Automated localization of language models, prefix models, word dictionaries etc.
- Lead memory optimization project that resulted in 30% memory saving.
- Developed C++ mapped memory tracking for Facebook scale.
Led software engineering team of 3 engineers, implemented agile methodology with bug tracker and git, practiced test-driven development. Worked with US client on the long term relationship basis. Some projects:
- Smoke detection video analytics, based on features detection on video. | C++, opencv
- Video analytics application server for 3-rd part analytics modules like car plate recognition, sensor line crossing, counting people in a queue. Server matches video streams with video analytics and provides frontend for configuration, handles full lifecycle | Java, Spring, JPA, Jetty, JS
- Security Entrance System with 3D face scanners for ArcelorMittal | Java, Spring MVC, JNI
- Frontend & middle layer for car plate recognition on Hikvision cameras | JS, lua
- Ware-tracking System with video monitoring | Python, Django, JS, HTML5, CSS
- NLP based classification service that matches question and category. | C#, Java, WordNet
- Debt collection agency phone number crawler | Ruby, PHP, GoogleSearch API
- Designed lead source tracking system and integrated SugarCRM with Google Analytics and ifbyphone | PHP, JS, CSS
- Completed more than 10 web development projects for website with 50 000 unique visitors per month.
