Description
We are a team of researchers and engineers developing post-training (SFT, RL) and controlled generation technologies for image generation, video, and omni-modal models. We are focused on improving the quality, controllability, and stability of generative models using RLHF / RLAIF methods and instructional editing. In addition to fundamental research, we create applied solutions for B2B partners, ranging from interior generation systems to personalized avatar generation systems.
Responsibilities
- designing and training high-performance diffusion models (R2V, R2I, R2V+A) for video/image editing and generation
- developing new approaches and architectural solutions for the post-training phase of diffusion models: RLHF (PPO, DPO, etc.), RLAIF, SFT
- developing, scaling, and maintaining RL pipelines
- leading experiments: hypothesis formulation, protocol development, results analysis
- conducting code reviews, mentoring junior researchers and engineers, participating in research roadmap planning
- collaborating with Pretraining, Data, Infrastructure, and Production teams to scale pipelines and enhance model stability in production
- maintaining and implementing SOTA approaches: monitoring literature (arxiv, NeurIPS, ICML, CVPR, ICLR), initiating internal research.
Requirements
- bachelor's/master's degree in computer science/applied mathematics/machine learning or related fields
- 3+ years of relevant research and development experience in deep learning/computer vision/generative AI
- deep knowledge in Computer Vision and Generative Modeling: Diffusion Models, GANs, VAEs, Flow/Rectified Flow Matching
- experience with modern diffusion frameworks (Diffusers) and models (FLUX, Wan 2.X, etc.)
- proficient in PyTorch and distributed training skills (DDP/FSDP)
- understanding and practical application of RL and RLHF ( PPO/DPO, etc.)
- ability to design architectures, plan experiments, and interpret results.
Conditions
- largest DS&AI community — over 600 DS specialists of the bank
- digest of the latest developments in the DS&AI field and reports from major world conferences
- opportunity to be a co-author of research papers and articles for international conferences
- opportunity to choose a convenient work format: hybrid or office
- annual salary review, annual bonus
- corporate gym and recreation areas
- more than 400 educational programs from SberUniversity for professional and career development
- extended DMS, preferential insurance for family and corporate pension program
- mortgage more profitable up to 7% for each employee
- free SberPrime+ subscription, discounts on products from partner companies
- referral bonus for recommending friends to the Sber team.
