Advisory Researcher - ML Engineer
Please note that the team has moved to a new location. The office is now at 121 Menachem Begin Road, Tel Aviv, 61st floor, in the POINT office complex. As per our office policy, you will be required to visit the site at least three times a week. Lenovo Digital Trust Lab is seeking a hands-on ML Engineer to join our Hybrid AI Security team. This role focuses on model build and fine-tuning workflows, working closely with security researchers to embed security and Responsible AI capabilities into the training lifecycle. You will work at the intersection of machine learning, security, and platform engineering—developing tooling, pipelines, and controls that help ensure models are trained responsibly, securely, and in alignment with Lenovo’s Trust, Privacy, and Responsible AI principles. Job Responsibilities: Design and maintain ML training and fine-tuning pipelines for LLMs and ML models. Work closely with security researchers to integrate security-driven requirements into data preparation, training, and build workflows. Implement ML-side capabilities needed to support security controls (e.g., data filtering hooks, dataset labeling, training instrumentation). Implement controls to detect and mitigate training-time threats (data poisoning, backdoors, contamination, leakage). Develop fine-tuning workflows (e.g., LoRA/PEFT) with auditability, reproducibility, and policy enforcement. Integrate model evaluation and testing frameworks focused on security, robustness, and safety into the build phase. Minimum Requirements: 3+ years of experience as an ML Engineer or Applied ML Engineer. Strong experience with Python and ML frameworks (PyTorch, TensorFlow, or JAX). Hands-on experience training or fine-tuning ML models or LLMs. Solid understanding of ML pipelines: data ingestion, preprocessing, training, evaluation. Familiarity with security and Responsible AI concepts (data integrity, bias, robustness). Preferred Requirements: Experience with LLM fine-tuning techniques (LoRA, QLoRA, PEFT, adapters). Familiarity with fine-tuning frameworks and tooling (Hugging Face, LLaMA Factory, Unsloth). Knowledge of training-time attack vectors (data poisoning, backdoor attacks, distribution shift). Experience implementing data scanning, validation, or quality checks in ML pipelines. Exposure to secure ML build practices (experiment tracking, versioning, lineage). Experience integrating evaluation tools (security, safety, or RAI) into CI/CD or ML pipelines. Familiarity with governance or compliance-driven ML workflows What Lenovo Can Offer You: Opportunities for career advancement and personal development Access to a diverse range of training programs Performance-based rewards that celebrate your achievements Flexibility with a hybrid work model (3:2)