Senior AI / LLM Engineer

Senior AI / LLM R&D Engineer

Location: Remote - Europe Time Zones

Level: Senior / Staff

Experience required: 5+ years in machine learning / deep learning engineering with direct hands-on work on large language models

About the role

We are seeking an exceptional Senior LLM R&D Engineer to join our frontier AI research & development team. You will work on next-generation large language models, post-training techniques, reasoning & agentic systems, efficient scaling, and multimodal capabilities.

Your work will directly contribute to training / post-training of models in the 100B–1T+ parameter range, pushing state-of-the-art performance in reasoning, long-context understanding, tool use, and safety/alignment.

This is a highly technical, research-oriented engineering role — ideal for people who love reading recent papers, implementing them at scale, and shipping improvements to production models.

Key Responsibilities

• Design, implement and scale pre-training, post-training (SFT, DPO, PPO, RLVR, online RL, etc.) and inference pipelines for very large language models
• Conduct cutting-edge experiments in model architecture, mixture-of-experts (MoE), state-space models, test-time compute scaling, long-context training, and multimodal integration
• Own end-to-end ownership of major R&D initiatives: from data curation → training recipe → evaluation → iteration
• Optimize training & inference efficiency (memory, throughput, energy) using latest distributed training techniques and hardware-aware optimizations
• Build robust evaluation harnesses (human & automated) for reasoning, agentic capabilities, safety, and factual accuracy
• Collaborate closely with researchers, systems engineers, data team, and product teams to turn research ideas into production-grade model improvements
• Stay at the absolute forefront of the field — implement ideas from recent ArXiv / conferences within days/weeks

Required Qualifications

• 5+ years of professional experience building, training or fine-tuning large-scale deep learning models (LLMs ≥ 13B strongly preferred)
• Very strong PyTorch proficiency (JAX bonus)
• Deep hands-on experience with modern LLM training & post-training stacks:
  • Hugging Face Transformers, Accelerate, PEFT (LoRA/QLoRA/DoRA/LoHa…)
  • DeepSpeed, Megatron-LM, vLLM, SGLang, TRL, Axolotl, Unsloth, Liger-kernel, FSDP2 / torch.compile
  • Alignment toolkits: TRL, Aligners, RLAIF pipelines, reward modeling
• Proven track record shipping LLM improvements in at least one of: reasoning, long-context, tool use / agents, multimodality, multilingual, or safety
• Strong software engineering skills: clean, modular, well-tested code; experience with large codebases
• Comfortable working at scale: multi-node (hundreds of GPUs), massive datasets (trillions of tokens), long-running jobs
• Solid understanding of latest advances in LLMs (2024–2026 papers): MoE scaling laws, test-time scaling, inference-time search, chain-of-thought distillation, RL for reasoning, etc.

Strongly Preferred / Differentiating Skills

• Experience training or heavily customizing models ≥ 70B parameters
• Publications (even workshop) or strong open-source contributions in LLM space
• Hands-on work with frontier open models (Llama-3/4, DeepSeek-R1/R2, GLM-4.x, Qwen-2.5, Mistral Large / Nemo, Gemma-2/3…)
• Expertise in any of: agent frameworks (LangGraph, CrewAI, AutoGen), RAG at scale, long-context architectures (RingAttention, Infini-Attention, Mamba-2 hybrids), speculative decoding, quantization (GPTQ/AWQ/FP8), multimodal (vision-language, video-language)
• Experience with very large-scale data pipelines (Spark, Ray Data, web-scale deduplication/filtering)
• Familiarity with modern evaluation suites (Arena-Hard, GPQA, SWE-bench Verified, FrontierMath, LiveCodeBench…)

What we offer

• Work on frontier-scale models with meaningful compute budget
• Direct impact on next-generation AI systems
• Competitive compensation package (base + meaningful equity + performance bonus)
• Top-tier compute resources and latest hardware
• Flat structure — your ideas matter regardless of title

If you're passionate about building the next generation of reasoning & agentic foundation models and have shipped serious LLM work in the past 5 years — we want to talk to you.

Apply with: CV + GitHub / Hugging Face profile + short note about your most impressive LLM project (training, post-training, inference optimization, agent system…)