Applied Research Scientist — Multimodal AI Agents

Prime Intellect, Inc.; Greater London; Remoto; GBP 113.000 - 225.000; Tempo pieno

Vantaggi offerti da questo lavoro

Visa Sponsorship & relocation support

Flexible work locations (London, SF,NY

Professional development budget

Team off-sites & conference attendance

Descrizione del lavoro

Prime Intellect, Inc. is building an open frontier AI stack and post-training evaluation systems. This Applied Research role focuses on turning real workflows into environments, rewards, and experiments to drive reliable improvements across browser, desktop, and cross-application agents.

You will work externally with customers and research partners while pursuing a strong research agenda. The role emphasizes designing end-to-end experiments, assessing whether gains come from models, data, or

Competenze

• Experience with multimodal agent systems and long-horizon tasks.
• Ability to build end-to-end experiments and evaluate improvements.
• Strong Python and tooling for model training and evaluation.

Mansioni

• Design and build multimodal environments from real workflows and tools.
• Develop browser, desktop, and cross-application agents with robust sandboxing.
• Run post-training experiments and compare model vs data-driven gains.
• Lead recovery and generalization experiments across apps, layouts, and permissions.

Conoscenze

Multimodal agents

Long-horizon evaluation

Browser automation

Desktop automation

Reinforcement learning

Imitation learning

Human–computer interaction

Experiment design

Python

Strumenti

Python

Descrizione del lavoro

Prime Intellect, Inc. is building an open frontier AI stack and post-training evaluation systems. This Applied Research role focuses on turning real workflows into environments, rewards, and experiments to drive reliable improvements across browser, desktop, and cross-application agents.

You will work externally with customers and research partners while pursuing a strong research agenda. The role emphasizes designing end-to-end experiments, assessing whether gains come from models, data, or