Job Description
About Toptal Toptal is a global network of top talent in business, design, and technology that enables companies to scale their teams, on-demand. With $200+ million in annual revenue and team members based around the globe , Toptal is the world’s largest fully remote workforce . We take the best elements of virtual teams and combine them with a support structure that encourages innovation, social interaction, and fun. We see no borders, move at a fast pace, and are never afraid to break the mold. Job Summary Toptal is building a dedicated AI Research team focused on advancing the frontier of agentic AI systems powered by proprietary real-world interaction data. We are seeking AI Researchers who are excited to explore how large-scale, real-world signals can be transformed into better reasoning, improved generalization, and more capable multimodal agents. In this role, you will work at the intersection of model development, multimodal representation learning, and reinforcement learning, designing new approaches that enable agents to learn from complex behavioral data, workflows, and multimodal inputs such as audio, logs, and structured interaction traces. You will focus on building and improving learning systems for agents, including methods for RAG, fine-tuning, reinforcement learning (RLHF, DPO, GRPO), and joint embedding spaces, as well as speech and audio intelligence capabilities such as STT, ASR, and audio signal modeling. You will collaborate closely with engineering and product teams to ensure research breakthroughs are translated into scalable systems, and that feedback from production continuously improves model behavior. This is a remote position. All communication and resumes must be in English. Responsibilities: The following information is intended to describe the general nature and level of work being performed. It is not intended to be an exhaustive list of all duties, responsibilities, or required skills. Advance research on agentic AI systems trained on real-world interaction signals and multimodal data. Design and experiment with learning paradigms for large-scale models, including RAG, supervised fine-tuning, RLHF, DPO, and GRPO-style methods. Develop multimodal representation learning approaches, including joint embedding spaces across text, audio, logs, and structured interaction traces. Improve speech and audio intelligence capabilities, including STT, ASR, and audio-driven learning signals. Research methods for enhancing agent reasoning, planning, tool use, and adaptation in real-world environments. Define how complex behavioral and interaction signals can be translated into effective training objectives for large-scale models. Build and refine evaluation methodologies for agent performance in real-world, domain-specific scenarios. Collaborate with engineering and product teams to bring research ideas into production systems. Identify patterns in real-world workflows and convert them into generalizable modeling and representation strategies. Contribute to the long-term research direction of Toptal’s agentic AI systems and multimodal capabilities. Stay current with academic and industry research and integrate relevant advancements into internal systems. In the first week, expect to: Join the AI team and orient yourself with Toptal’s mission and strategy. Access our existing datasets, agent stacks, and internal evaluation tools. Map the landscape of raw data sources currently feeding our agentic systems. In the first month, expect to: Develop a deep understanding of our current architectures and evaluation methodologies. Identify high-leverage gaps where data improvements can measurably increase agent capability. Initiate concrete improvements to pipelines converting raw inputs into model-ready assets. Shape feedback loops that utilize live performance as a training signal. In the first three months, expect to: Own a production data pipeline from ingestion through delivery into RL or fine-tuning workflows. Define reusable schemas that abstract repeated workflows into queryable formats. Drive measurable advancements in agent accuracy within a specific vertical, backed by metrics. Integrate AI features into user-facing surfaces like browsers or enterprise tools. In the first six months, expect to: Lead the design of multimodal pipelines that unify text and real-time logs for agents. Establish tooling for encoding institutional knowledge into scalable schemas for the team. Define the team’s strategy for fine-tuning and capturing human feedback for RLHF. Mentor teammates on data-centric approaches and influence the team’s technical direction. In the first year, expect to: Serve as a key technical leader in turning proprietary data into a durable competitive advantage. Operate as a recognized expert across the team on knowledge representation and improvement loops. Drive a step-change in agent capability across multiple verticals through clear performance metrics. Shape the next generation of
