Applied Machine Learning Engineer, EMEA

Excerpt from the original posting

About Us: 
Fireworks is the platform for specialized intelligence, enabling companies to build, train, and serve AI models tailored to their own data, workflows, and products. Founded by the team behind PyTorch and backed by AMD, Atreides, Benchmark Capital, Index Ventures, Lightspeed, NVIDIA, Sequoia Capital, and TCV, Fireworks powers production AI with hundreds of state-of-the-art open models across text, image, embedding, audio, and multimodal workloads. Today, Fireworks is a Series D company valued at $17.5 billion, bringing together an ambitious, collaborative team that's building the future of enterprise AI.
The Role: 
As an Applied Machine Learning Engineer, you are the technical owner of a customer engagement. This is what is globally known as the Forward Deployed Engineer role; at Fireworks in EMEA it carries the AMLE title. You embed with a small number of accounts, work in their environment, and own whatever stands between them and production on Fireworks: the model, the serving path, the integration, the performance. You scope the work before signature and deliver what you scoped. Depth over breadth: few accounts, owned end to end.
What we value most is first principles thinking: reasoning about unfamiliar problems from the ground up rather than from playbooks. You will be evaluated on this directly in the interview process, and strength here can outweigh gaps elsewhere in your background.

Four Spikes: 
- Software engineering. You ship production systems and are comfortable dropped into an unfamiliar codebase or a customer environment with a deadline. Backend and systems depth, strong Python plus at least one systems language, and fluency with AI-assisted and agentic engineering as a core part of how you build.

- Machine learning. You take open models and make them better on a customer's task: SFT, LoRA and other parameter-efficient methods, reinforcement learning (RLHF, RLVR), distillation. Just as important, you design evaluations that reflect the real task and let them drive training decisions, and you know when fine-tuning is not the answer.

- Infrastructure and performance. You know your GPUs: memory bandwidth, kernels, parallelism, what utilization numbers actually mean. Serving and inference optimization: quantization, speculative decoding, batching, KV cache behavior, latency vs throughput tradeoffs, multi-GPU and multi-node serving, capacity planning. You can take a deployment that works and make it fast and economical, and reason a…

→ Fireworks · Ashby