LLM Research & Validation Specialist

About the role

The LLM Research & Validation Specialist will lead frontier research, quantitative evaluation and independent validation of large language models, multimodal models, retrieval‑augmented generation systems and agentic AI used or proposed by ADIB. The role translates mathematical and scientific methods into reproducible tests and provides decision‑useful evidence for model governance.

Key responsibilities

• Lead independent validation of LLM, multimodal, RAG and agentic AI use cases across design, implementation, deployment and monitoring.
• Design reproducible evaluation harnesses, golden datasets, adversarial suites, counterfactual tests and statistically defensible acceptance criteria.
• Assess transformer architecture components such as tokenisation, embeddings, attention, decoding, fine‑tuning, alignment and inference configuration.
• Perform deep testing of prompt injection, data leakage, tool‑use safety, excessive agency and multi‑step failure propagation.
• Build and maintain reusable Python‑based validation tooling, automated test pipelines, experiment tracking and technical documentation.
• Prepare validation reports, research notes and senior‑management briefings that clearly distinguish evidence, judgement and residual uncertainty.

Required profile

• Master's degree in a quantitative discipline (Physics, Mathematics, etc.); PhD or research‑intensive master’s strongly preferred.
• 1–3 years of experience in AI research, machine learning, quantitative modelling, model validation or scientific computing.
• Deep understanding of probability, statistics, linear algebra, optimisation, numerical computation, experimental design and uncertainty quantification.
• Strong knowledge of transformers, LLM training and inference, embeddings, RAG, fine‑tuning, alignment and AI safety failure modes.

Required skills

• Advanced Python programming.
• Experience with PyTorch and Hugging Face libraries.
• Proficiency in SQL and Git.
• Familiarity with cloud AI platforms and experiment‑tracking frameworks.