For one of our clients, we are looking for a Senior LLM Researcher - Alignment to help build reliable, safe, and high-performing language models.
In this role, you will design and implement alignment, sanitization, evaluation, and guardrailing pipelines that enable the delivery of trustworthy AI at scale. You will also play a key role in advancing Arabic-focused language model capabilities, including dialectal coverage and culturally relevant safety considerations.
Key Responsibilities
Own alignment strategies, including:
Supervised fine-tuning (SFT)
Preference optimization, such as DPO and RLHF
Reward modeling
Constrained decoding
Design and build end-to-end alignment pipelines covering data curation, training, preference optimization, and evaluation.
Develop data pipelines for filtering, augmentation, preference-data collection, and quality control.
Build model sanitization workflows, including:
PII removal
Data redaction
Prompt and response scrubbing
Implement runtime guardrails using policy models, classifiers, rule engines, retrieval-based safety checks, and rate controls.
Establish red-teaming and adversarial testing programs, including incident taxonomies and response playbooks.
Define and maintain evaluation frameworks covering safety, robustness, task performance, drift, latency, and telemetry.
Expand Arabic language coverage across dialects, domains, and culturally relevant safety scenarios.
Partner with product, platform, and engineering teams to move research into production.
Mentor engineers and researchers while developing internal documentation, playbooks, and best practices.
Lead major AI initiatives that improve model helpfulness, harmlessness, faithfulness, reliability, and cost efficiency.
Requirements
Proven research or industry experience training and aligning large language models.
Strong hands-on experience with SFT and preference optimization, including RLHF, DPO, or reward modeling.
Strong machine learning theory and engineering capabilities.
Demonstrated experience in LLM safety and security, including guardrails, classifiers, adversarial prompting, or PII handling.
Experience designing evaluation frameworks for model quality, safety, robustness, and reliability.
Track record of taking research from experimentation through production, supported by clear metrics and documentation.
Strong programming skills in Python and experience with modern deep learning frameworks, particularly PyTorch.
Ability to work cross-functionally and communicate complex research clearly.
Experience mentoring engineers, researchers, or technical teams.
Nice to Have
Experience with red-teaming, AI policy design, or compliance in multilingual environments.
Familiarity with Arabic NLP, including dialectal variation.
Knowledge of retrieval-augmented generation, constrained decoding, or tool-use agents.
Experience with inference optimization, quantization, low-latency serving, or observability.
Familiarity with distributed training environments and large-scale model deployment.
Experience with systems such as Hugging Face, Slurm, LlamaFactory, NeMo, RLHF/DPO frameworks, vector databases, and policy or guardrail services.
Experience working with H200 or similar GPU infrastructure.
#J-18808-Ljbffr
Senior LLM Researcher - Alignment in London employer: Astek
Join a forward-thinking company that values innovation and collaboration, offering a dynamic work environment in either London or Riyadh. As a Software Architect, you will benefit from a culture that prioritises professional growth, with opportunities to lead cutting-edge projects in robotics while enjoying competitive benefits and a commitment to work-life balance. Embrace the chance to be part of a team that champions high-quality engineering practices and fosters continuous learning.
