AI Researcher (Remote)

About the role

We are building a dedicated AI Research team at Toptal to push the frontier of agentic AI systems that learn from real‑world interaction data. As an AI Researcher you will design and experiment with novel learning paradigms that combine multimodal signals such as text, audio, logs and structured traces.

Key responsibilities

• Advance research on agentic AI systems trained on real‑world interaction signals and multimodal data.
• Design and experiment with large‑scale learning paradigms including Retrieval‑Augmented Generation (RAG), supervised fine‑tuning, RLHF, DPO and GRPO‑style methods.
• Develop multimodal representation learning approaches and joint embedding spaces across text, audio, logs and structured interaction traces.
• Improve speech and audio intelligence capabilities such as speech‑to‑text (STT), automatic speech recognition (ASR) and audio‑driven modeling.
• Research methods to enhance agent reasoning, planning, tool use and adaptation in real‑world environments.
• Define training objectives that translate complex behavioral and interaction signals into effective large‑scale model learning.
• Build and refine evaluation methodologies for agent performance in domain‑specific scenarios.
• Collaborate with engineering and product teams to translate research breakthroughs into scalable production systems.

Required profile

• Strong background in AI research with experience in model development, multimodal representation learning and reinforcement learning.
• Ability to work remotely and communicate fluently in English.
• Proven track record of collaborating with engineering and product teams to ship research‑driven features.

Required skills

• Retrieval‑Augmented Generation (RAG)
• Supervised fine‑tuning
• Reinforcement Learning from Human Feedback (RLHF)
• Direct Preference Optimization (DPO)
• Generalized Reinforcement Preference Optimization (GRPO)
• Multimodal representation learning and joint embedding spaces
• Speech‑to‑text (STT) and Automatic Speech Recognition (ASR)
• Audio signal modeling