Research Scientist

Salesforce, Inc..; Greater London; On-site; GBP 70,000 - 100,000; Full time

Job summary

An innovative technology company in London is seeking Members of Technical Staff to lead research initiatives on training foundation models for enterprise agents. Ideal candidates will have a strong academic background, deep knowledge of transformer architectures, and proficiency in PyTorch. Responsibilities include designing fine-tuning techniques and developing synthetic data generation methods. An excellent opportunity to contribute to cutting-edge research in AI, with potential for publication in top-tier conferences.

Qualifications

• Strong theoretical understanding of deep learning fundamentals.
• Expertise in fine‑tuning open source models using reinforcement learning.
• Research experience in top-tier ML conferences or journals.

Responsibilities

• Design and implement novel fine‑tuning techniques.
• Develop synthetic data generation pipelines.
• Optimize model performance through rigorous experimentation.

Skills

Deep knowledge of transformer architectures

Proficiency in PyTorch

Experience with distributed training

Education

PhD in Computer Science or related field

Tools

DeepSpeed

Ray

Job description

About Us

In June 2025, Convergence AI was acquired by Salesforce, joining the Agentforce team. The goal is to establish Salesforce's first engineering team in London, focusing on agent research topics and developing the next generation of personal assistants capable of automating large, previously labor-intensive workstreams for the enterprise world.

The Role

We are looking for Members of Technical Staff with strong academic backgrounds to join our team and lead research initiatives on training foundation models for a brand new enterprise agent. You will work with a small team of hands‑on researchers — equipped with substantial GPU resources – to advance the state‑of‑the‑art in multi‑modal vision‑language models, reinforcement learning, and action models. You will develop novel architectures and training approaches while publishing your findings in top‑tier AI conferences and journals.

Responsibilities

Your role will span both theoretical research and practical implementation:

• Designing and implementing novel supervised fine‑tuning, preference learning and reinforcement learning techniques
• Developing innovative methods for data curation, including synthetic data generation pipelines and optimal dataset composition strategies
• Conducting rigorous experimentation to optimize model performance through data mixes, regularization techniques, and low‑level optimizations
• Advancing the theoretical understanding of transformer architectures and their applications to multi‑modal learning
• Publishing research findings in top‑tier venues such as NeurIPS, ICML, and ICLR

Requirements

• Deep knowledge of transformer architectures and vision‑language models and strong theoretical understanding of deep learning fundamentals
• Expertise in training and fine‑tuning open source models using techniques such as supervised fine‑tuning or reinforcement learning
• Proficiency in PyTorch and related frameworks
• Experience with large‑scale distributed training and inference using e.g. DeepSpeed, FSDP, Ray

Bonus Qualifications

• PhD in Computer Science, Machine Learning, or related field
• Strong publication record in top‑tier ML conferences (NeurIPS, ICML, ICLR) or top journals
• Research experience at leading academic or industrial research labs
• Demonstrated expertise in reinforcement learning
• Contributions to open‑source ML frameworks
• Experience developing novel datasets or data generation approaches
• Background in causal reasoning or alignment research