#### About the Role
We're looking for a Machine Learning Engineer to fine-tune open-source LLMs and deploy them as fast, cost-efficient production services. You'll own the full path from dataset preparation and training through quantisation, serving and performance optimisation.

#### Key Responsibilities
- Fine-tune open-source LLMs and SLMs (e.g., Llama, Qwen, Mistral, Gemma, DeepSeek) using SFT, LoRA/QLoRA and other PEFT techniques
- Build, clean and validate domain-specific instruction datasets
- Configure and run GPU training, including hyperparameters, mixed precision (BF16/FP16), gradient accumulation and checkpointing
- Evaluate fine-tuned models against baselines using measurable criteria
- Quantise and optimise models for production (AWQ/GPTQ, INT8/INT4, KV cache, continuous batching)
- Deploy and serve self-hosted models with vLLM, TGI, Triton or similar
- Benchmark and improve latency, throughput, GPU utilisation and inference cost
- Expose models through production APIs (FastAPI or similar)

#### Must-Haves
- Strong Python and PyTorch
- Hands-on experience fine-tuning LLMs with Hugging Face Transformers, PEFT and TRL (or Axolotl/Unsloth)
- Solid grasp of transformer architecture and training fundamentals
- Experience deploying LLMs on GPU infrastructure with vLLM, TGI, Triton or equivalent
- Practical experience with quantisation and inference optimisation
- Experience with Docker and Linux environments

#### Nice-to-Have
- RAG, vector databases and AI agent development
- Automated evaluation pipelines and LLM metrics (hallucination, faithfulness)
- MLOps tooling: MLflow, W&B, Kubernetes, CI/CD for ML
- AWS: SageMaker, EKS, Bedrock
- Distributed training (DeepSpeed, multi-GPU), knowledge distillation, CUDA fundamentals
- Experience with A100/H100/L40S GPUs