AI Researcher (Member of Technical Staff)

Perplexity AI; Berlin; Vor Ort; EUR 90.000 - 130.000; Vollzeit

Zusammenfassung

Perplexity AI in Berlin seeks top-tier AI Research Scientists and Engineers to advance our AI products and capabilities. You will contribute to Sonar models, Deep Research Agent, Comet Agent, and Search products, building SOTA experiences at scale.

Join one of three teams: Core Research for base models, Agent Products for Deep Research Agent and Labs/Canvas, and Comet Agent for specialized use cases. Post-train LLMs, develop training infra, and collaborate to deliver user experiences.

Qualifikationen

• PhD in ML/AI or related field.
• Experience with large-scale LLMs.
• Strong publications and research impact.

Aufgaben

• Post-train SOTA LLMs using supervised and reinforcement learning techniques.
• Leverage our dataset to scale model performance across Sonar, Deep Research, Comet, and Search.
• Stay current with LLM research, especially in model training, optimization, and personalization techniques.
• Implement preference optimization and personalization capabilities to enhance user experience.
• Invent in-house improvements and optimizations to enhance SOTA models.
• Turn research ideas into algorithms and run experiments to launch new models.
• Own full-stack data, training, and evaluation pipelines for model development.
• Build robust training frameworks (Megatron/PyTorch) for post-training LLMs.
• Integrate models into Perplexity’s product suite and work with engineering teams.

Kenntnisse

Python

PyTorch

Large-scale LLMs

Reinforcement learning

Post-training techniques

C++

CUDA

Agent systems

Multi-step reasoning

Personalization

Publications

Ausbildung

PhD in Machine Learning / AI

Tools

Megatron-LM

Jobbeschreibung

Perplexity is seeking top-tier AI Research Scientists and Engineers to advance our AI products and capabilities. We’re building the future of AI-powered search and agent experiences through our Sonar models, Deep Research Agent, Comet Agent, and Search products. Join us in creating SOTA experiences that handle hundreds of millions of queries and continue to scale rapidly. Depending on your interests and expertise, you’ll work on one of three specialized teams:

Core Research Team (Horizontal)

Focus on generating and improving base models that power all our products. This team works on foundational model capabilities, post-training techniques, building RL infra and infrastructure that benefits the entire organization

Agent Products Team (Vertical)

Concentrate on fine-tuning and optimizing models for our Deep Research Agent and Labs/Canvas products. This team bridges research and product, ensuring our agent capabilities deliver exceptional user experiences

Comet Agent Team (Vertical)

Dedicated to developing and enhancing our Comet Agent product. This specialized team focuses on the unique requirements and optimizations needed for Comet’s specific use cases

Responsibilities

• Post-train SOTA LLMs using the latest supervised and reinforcement learning techniques (SFT/DPO/GRPO)
• Leverage our rich query/answer dataset to scale model performance across Sonar, Deep Research, Comet, and Search products
• Stay current with the latest LLM research, especially in model training, optimization, and personalization techniques
• Implement preference optimization and personalization capabilities to enhance user experience
• Invent in-house improvements and optimizations to enhance SOTA models
• Turn research ideas into algorithms and run experiments to launch new models
• Own full-stack data, training, and evaluation pipelines required for model development
• Build robust and effective training frameworks (on top of Megatron/PyTorch) for post-training LLMs
• Implement necessary infrastructure and components to support cutting-edge model training at scale
• Integrate models seamlessly into our product ecosystem
• Work closely with engineering teams to integrate models into Perplexity’s product suite
• Collaborate across teams to ensure cohesive AI experiences throughout our platform
• Partner with product teams to understand user needs and translate them into model improvements

Requirements

• Minimum 2-6 years of experience on relevant projects (depending on seniority level)
• Passion for tackling challenging problems
• Proven experience with large-scale LLMs and Deep Learning systems
• Experience with post-training techniques and reinforcement learning
• Strong programming skills in Python/PyTorch; versatility is a plus
• Self-starter with a willingness to take ownership of tasks
• PhD in Machine Learning, AI, Systems, or related areas
• Experience in post-training LLMs with SFT/DPO/GRPO
• C++/CUDA programming skills
• Experience building LLM training frameworks
• Experience with agent systems and multi-step reasoning
• Academic publications and research impact
• Background in personalization and preference learning