Description
We strive to develop artificial intelligence, understanding that building a safe and reliable platform is the foundation of our mission and customer trust. We are looking for an engineer who will help us set new standards in the field of Trust & Safety.
If you resonate with the spirit of Bell Labs or Xerox PARC in their best years — the spirit of deep research, hypothesis testing, and bringing ideas to a working product (production) — you are the one for us.
You will work at the intersection of research and engineering, creating systems to protect against risks associated with large language models (LLMs) and AI agents.
Responsibilities
- developing systems to prevent factual and contextual hallucinations, detect toxicity, discrimination, and malicious content
- developing Guardrails
- designing and implementing architectural solutions and tactics for model output restriction (guardrails) to protect users from abuse
- training models, fine-tuning and training multimodal models, creating new neural network architectures for safety tasks
- monitoring and response, identifying and resolving active incidents, creating tools and infrastructure for root cause analysis
- research and development, close collaboration with researchers to implement advanced AI Safety practices, participation in industry conferences, studying papers and adapting global experience for our products
- compliance assessment, measuring and improving the alignment of AI model and agent behavior with human values and ethical standards.
Requirements
- experience in security; you have experience in content moderation, fraud prevention, or abuse prevention
- excellent knowledge of Python, willingness to quickly dive deep into the Python ecosystem for ML
- experience deploying classifiers and ML models in production or a strong desire to master modern ML infrastructure
- understanding of the principles of designing and training neural networks: from classic CNN/RNN to modern transformers and multimodal architectures
- experience developing AI agents and understanding of Multi-Agent Systems architectures
- successful experience creating and supporting high-load services under rapid scaling conditions
- skill in critically assessing the risks of new features and finding innovative solutions to mitigate them without degrading the user experience (UX).
Conditions
- comfortable modern office near Kutuzovskaya metro station, hybrid work schedule, possibility to work remotely from another region
- annual salary review and performance-based annual bonus
- extended VHI, preferential insurance for family, corporate gym and relaxation areas
- access to more than 400 educational programs of SberUniversity for professional and career growth
- financial benefits: flexible mortgage discount (1/3 of the Central Bank's key rate), free SberPrime+ subscription, discounts from partners
- referral bonus for recommending friends to Sber teams.
