Senior LLM Researcher - Alignment

We are looking for a Senior LLM Researcher to help build reliable, safe and high-performance language models, with a particular focus on Arabic and multilingual AI. You will design and implement LLM alignment and sanitization pipelines, develop robust safety guardrails, establish rigorous evaluation frameworks, and help translate AI research into production-ready capabilities. This is an opportunity to work at the intersection of LLM research, AI safety, alignment and production engineering, contributing to high-impact AI initiatives and shaping the development of trustworthy language models. What You'll Do LLM Alignment & Post-Training

• Design and implement end-to-end LLM alignment pipelines covering data curation, supervised fine-tuning (SFT), preference optimization and rigorous evaluation.
• Apply alignment techniques including DPO, RLHF, reward modelling, rejection sampling and constrained decoding.
• Develop and improve preference datasets, including data collection, filtering, augmentation and quality control.
• Establish reproducible research processes, baselines and measurable performance metrics. Model Safety, Sanitization & Guardrails
• Build model sanitization pipelines covering PII removal, data redaction, prompt/response scrubbing and harmful-pattern filtering.
• Design and integrate runtime safety mechanisms including policy models, classifiers, rule engines and retrieval-based safety checks.
• Develop and maintain red-teaming and adversarial testing processes.
• Establish incident taxonomies and response playbooks to identify and address model safety risks. Evaluation & Reliability
• Build automated evaluation frameworks covering task performance, safety, robustness and model behaviour.
• Develop continuous testing and reporting integrated into CI and model development workflows.
• Monitor model quality, safety, drift and telemetry.
• Support A/B testing and canary releases and measure improvements in helpfulness, harmlessness and faithfulness. Arabic & Multilingual AI
• Improve LLM alignment for Arabic language use cases, dialectal coverage and culturally relevant safety considerations.
• Contribute to the development of preference data and evaluation methodologies appropriate for Arabic and multilingual models.
• Help identify and address language-specific safety and alignment challenges. Research, Leadership & Collaboration
• Stay current with research and best practices in LLM alignment, safety and post-training.
• Translate research findings into production-ready solutions.
• Partner with product and platform teams to deploy aligned models and measure real-world impact.
• Lead major AI initiatives and contribute to the AI roadmap.
• Mentor engineers and researchers on alignment development, evaluation and safety engineering.
• Create internal documentation, playbooks, experiments and best-practice guidelines. What We're Looking For Required
• Proven research or industry experience working on LLM training, post-training or alignment.
• Hands-on experience with Supervised Fine-Tuning (SFT) and at least one preference-optimization approach such as DPO or RLHF.
• Strong understanding of ML theory and practical ML engineering.
• Demonstrated experience in LLM safety, security or responsible AI, including guardrails, classifiers, adversarial prompting, red-teaming or PII handling.
• Experience designing or implementing LLM evaluation frameworks and measuring model performance, safety or robustness.
• Track record of translating research or experimentation into production-ready AI systems.
• Strong Python and PyTorch experience.
• Experience with modern LLM tooling and frameworks, particularly Hugging Face. Nice to Have
• Experience with Arabic NLP, Arabic language models or multilingual LLMs.
• Experience with red-teaming, AI safety or policy design in multilingual environments.
• Knowledge of RAG, constrained decoding and tool-use/agentic systems.
• Experience with inference optimisation, quantization, low-latency model serving or observability.
• Experience with distributed model training environments such as Slurm and frameworks such as LlamaFactory or NeMo.
• Experience working with preference-data and alignment tooling. Technical Environment Python | Rust | PyTorch | Hugging Face | SFT | RLHF | DPO | Reward Models | Slurm | LlamaFactory | NeMo | LLM Evaluation | Red Teaming | Guardrails | RAG | Vector Databases | Optimised Inference | Observability Why Join?
• Work on challenging LLM alignment and AI safety problems.
• Help shape the development of Arabic-oriented and multilingual AI.
• Work across research, engineering and product teams.
• Lead high-impact AI initiatives and influence technical direction.
• Contribute to building trustworthy, reliable and production-ready AI systems. Please note: This role is based in London and requires the successful candidate to work from the London office.