AI Engineer Azure AI / LLM / RAG

Your main responsibility will be to own the cloud fine-tuning of open-source LLMs (e.g., Llama 3.x 8B) on Azure - delivering both Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) end-to-end: data, training, evaluation, and deployment.

This role is hybrid, as onsite presence in the Valencia office will be required.

You will be responsible for:

• Run SFT on Azure AI Foundry (serverless fine-tuning) and Azure Machine Learning (managed GPU compute).
• Implement training with TRL SFTTrainer + PEFT (LoRA / QLoRA), bitsandbytes 4-bit quantization, and the Hugging Face stack (transformers, datasets, accelerate).
• Prepare and validate training data: chat-format JSONL, curation, deduplication, leakage-safe train/valid/test splits.
• Scale training with DeepSpeed / multi-GPU and optimize GPU memory and throughput.
• Design and run RFT on Azure AI Foundry: author and tune graders (string check, text similarity, custom Python, model-as-judge, multigrader).
• Engineer reward functions for verifiable tasks and mitigate reward hacking.
• Apply RL-based methods (RFT / GRPO / PPO, and preference methods such as DPO).
• Interpret reward curves and auto-evaluations; select checkpoints and tune RFT hyperparameters (reasoning effort, eval interval, compute multiplier).
• Build a rigorous evaluation harness: held-out test sets, task metrics, LLM-as-judge, and base-vs-fine-tuned regression checks.
• Deploy fine-tuned models to serverless / managed online endpoints; handle adapter merging, quantization, and inference serving.
• Own MLOps: reproducible Azure ML jobs, experiment tracking, dataset/model versioning, cost and quota management, and CI/CD.

• Bachelor's Degree in Information Technology, Cybersecurity, Computer Science, or related discipline-or equivalent professional experience.
• 5+ years engineering experience, incl. 2+ years hands-on LLM fine-tuning.
• Strong Python and PyTorch.
• Hugging Face transformers, peft, trl, datasets, accelerate.
• LoRA / QLoRA and parameter-efficient fine-tuning; bitsandbytes 4-bit quantization.
• Production experience on Azure Machine Learning and Azure AI Foundry (managed compute, serverless fine-tuning, online endpoints, GPU quotas).
• SFT delivery plus working knowledge of RL/preference methods (RFT, GRPO/PPO, DPO) and reward / grader design.
• Training-data engineering: JSONL/chat formats, curation, dedup, split hygiene, leakage prevention.
• Evaluation rigor: task metrics, LLM-as-judge, and reading training/validation curves.
• Linux, Git, containers, and cloud GPU workflows.
• DeepSpeed and distributed / multi-node training.
• Azure OpenAI fine-tuning API and the OpenAI-compatible SDK.
• RAG / GraphRAG, knowledge graphs, and synthetic training-data generation.
• Infrastructure-as-code (Bicep / Terraform), DevSecOps, and Azure cost optimization.
• Fluency in English (written and spoken).