Dzmitry Bahdanau is a research-driven software engineer with 15 years of experience bridging deep learning research and production-grade ML systems, currently based in Montreal and serving as a Team Member at Periodic Labs. He has led conversational AI research programs at ServiceNow Research and holds adjunct faculty ties at McGill and a core affiliation with Mila, combining academic rigor with industry impact. His work spans goal-driven dialogue, semantic parsing, AI-based code modeling, and efforts to reconnect symbolic AI with deep learning, often focusing on systematic generalization. A seasoned open-source contributor, he has improved core ML tooling and RL starter frameworks (e.g., adding gradient-norm monitoring and PPO support) and enhanced data pipeline robustness in widely used projects like fuel and BabyAI. Notably, Dzmitry blends hands-on backend engineering experience from Yandex-era search ranking and transliteration classifiers with contemporary research leadership, enabling practical solutions that are informed by deep theoretical insight.
15 years of coding experience
3 years of employment as a software developer
Bachelor's Degree Computer Science, Bachelor's Degree Computer Science at Belarusian State University
Master's Degree Computer Science, Master's Degree Computer Science at Jacobs University Bremen
BabyAI platform. A testbed for training agents to understand and execute language commands.
Role in this project:
Back-end Developer
Contributions:267 commits, 27 PRs, 91 pushes in 1 year 8 months
Contributions summary:Dzmitry contributed to the BabyAI platform by implementing features and improving the existing codebase. They introduced a default Proximal Policy Optimization (PPO) algorithm, created a script for printing instructions, and added logging for git information. Furthermore, the user modified the training script to print true rewards and enhanced the logging with a focus on training metrics.
Contributions:3 releases, 60 commits, 54 PRs in 2 years 6 months
Contributions summary:Dzmitry primarily contributed to the `fuel` data pipeline framework, focusing on enhancing the core functionality. They fixed bugs in the `Merge` transformer, introduced support for iteration schemes, and improved the handling of versioning. Furthermore, they added support for the `get_example_stream` method in the `H5PYDataset` class and made other improvements related to data handling and iteration.
data-pipelinemachine-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.