Machine Learning Engineer - LLMs

Who We Are You've ordered dinner through Just Eat Takeaway. You've bought and sold on OLX. You've paid with PayU. Chances are, you've already used a Prosus company today. We are Prosus, a global technology company powering some of the world's leading lifestyle ecommerce businesses. Every day, billions of people across India, Latin America, Europe and beyond use our products and services to buy, sell, pay, connect and discover. Our AI-driven ecosystem spans 100+ companies across ecommerce, food delivery, fintech and experiences, serving billions of consumers worldwide. AI powers how we operate and innovate, with more than 60,000 active AI agents helping our teams build, scale and move faster. Through our investment team, we also back the founders and technologies shaping the future: From robotics and autonomous systems to synthetic biology and frontier AI. Guided by our five values Innovation, Results, People, Impact, and Entrepreneurship, culture is our competitive advantage. We thrive with experimentation; we build, we test, we learn, we iterate. Join us in shaping the future of technology at global scale. Join our AI team at Prosus, the largest consumer internet company in Europe and one of the biggest tech investors in the world. You'll be working on the team that drives growth and innovation across the company, with your work directly impacting how millions of people shop online. WHO WE'RE LOOKING FOR We're seeking an experienced Machine Learning Engineer to develop our domain-specific language models. You'll own the full training pipeline - from continued pre-training on proprietary data, through fine-tuning, to advanced post-training techniques. You have hands-on experience training language models at scale (from embedding models to specialized LLMs), deep understanding of distributed systems and excel at turning research concepts into robust production systems. You're motivated by seeing your work deployed and used by millions of real users. You thrive in fast-paced environments where you balance research insights with practical business impact. WHAT YOU'LL DO - Analyze model performance and training data to identify improvement opportunities and formulate hypotheses for enhancement - Design, execute, and rigorously evaluate experiments to improve model quality, training efficiency, and downstream task performance - Train large language models for our domain through continued pre-training and full parameter fine-tuning on proprietary datasets - Design and implement post-training pipelines using RLHF, DPO, GRPO, and other reinforcement learning techniques - Build and optimize distributed training infrastructure across multi-node GPU clusters using frameworks like DeepSpeed, FSDP, Megatron-LM, or Axolotl - Own large-scale data preparation: filtering, quality assessment, deduplication, and data mixture strategies for training corpora at 100B+ token scale - Generate and curate high-quality synthetic data for instruction fine-tuning and capability enhancement - Debug training stability issues, optimize training throughput, and monitor model performance throughout long-running distributed jobs - Implement techniques from recent papers and adapt them to our use cases - Build robust evaluation frameworks to measure model quality and guide training decisions - Write production-grade, well-tested code and contribute to team engineering standards MINIMUM QUALIFICATIONS - 3+ years of ML engineering experience - Proven experience training and deploying language models to production (embedding models, encoder models, or large language models) including pre-training, continued pre-training, or fine-tuning with rigorous evaluation - Experience preparing large-scale training datasets: data filtering, quality assessment, deduplication strategies, and data mixture design - Hands-on experience with distributed training frameworks (DeepSpeed, FSDP, Megatron-LM, or Axolotl) including orchestrating multi-node jobs, debugging failures, and optimizing throughput - Strong understanding of training dynamics at scale: debugging loss instabilities, tuning learning rate schedules, managing training stability across long-running multi-node jobs - Expert Python and PyTorch with production experience using training libraries (Transformers, DeepSpeed, Accelerate) - Strong software engineering practices: version control, testing, CI/CD, code review PREFERRED QUALIFICATIONS - Published research at ML conferences (NeurIPS, ICML, ICLR, ACL, EMNLP), released models on Hugging Face, created public benchmarks, or contributed to open-source projects - Experience with post-training methods: RLHF, DPO, GRPO, or other reinforcement learning approaches for alignment and instruction-following - Experience optimizing models for production inference: quantization, model compression, distillation, and serving frameworks (vLLM) - Understanding of memory optimization: gradient checkpointing, mixed precision training (FP16, BF16, FP8), ZeRO optimization - Track record of building synthetic data generation pipelines for instruction tuning or domain adaptation - Research contributions: published papers at ML conferences, released models on Hugging Face, or created benchmarks used by the community - Contributions to open-source LLM training projects or tools WHAT WE OFFER - High-impact AI projects that are strategically vital to the company, with direct engagement from senior leadership including the CEO - State-of-the-art infrastructure: H200 GPU fleet, massive proprietary datasets, access to frontier models (OpenAI, Anthropic, Google, Together.ai http://Together.ai) for evaluation and baselines - Expert colleagues who have released top Hugging Face models, authored papers at NeurIPS, created well-known benchmarks, and built multiple production AI systems - Significant autonomy and freedom to test ideas, experiment with new approaches, and drive technical decisions - Modern tooling: Latest ML frameworks,