Forward Deployed Engineer (Model Customization and Optimization), AWS Forward Deployed Engineering

As a Forward Deployed Engineer II on the Custom Model and Optimization team, you embed with strategic customers to adapt and run custom generative AI models in production. You own end-to-end model workstreams—from data prep and evaluation to training, optimization, and deployment—tackling high-impact ML challenges. You collaborate with customer teams, shape technical plans, and translate business goals into measurable success metrics. You’ll contribute to scalable tooling and share learnings across AWS.

• Work/Life Balance
• Inclusive Team Culture
• Mentorship and Career Growth

• Embed within customer ML and engineering teams to understand business, data, model stack, and constraints
• Translate goals into measurable evaluation bar and technical plan (model selection, customization, serving architecture)
• Customize post-train models via continued pre-training, SFT, RLHF, and other RL methods
• Build data pipelines, synthetic data, and evaluation harnesses to prove model improvement over baselines
• Design and implement distributed training and RL post-training pipelines using frameworks (PyTorch FSDP2, Megatron-LM, Miles, verl)
• Develop production inference stacks and optimize throughput, latency, memory, and cost per token (SGLang, vLLM; quantization, batching)
• Migrate and tune workloads to AWS Trainium with AWS Neuron SDK; profile and close gaps vs GPU baselines
• Codify reusable fine-tuning recipes, evaluation frameworks, benchmarking tools, and patterns
• Share field learnings with AWS service teams (Bedrock, SageMaker AI, Neuron); identify expansion opportunities

• 3+ years of non-internship software development experience
• Experience programming in at least one modern language (Java, C++, or C#)
• 2+ years in an Agile, startup, or fast-scaling environment
• Hands-on experience with generative AI tech (LLMs, vision-language, or speech models)
• Experience with AI tools: eval frameworks, agent tooling, RAG pipelines, and prompt engineering

• strong communication of technical concepts to both technical and non-technical audiences
• autonomous, proactive contributor
• coaching others on code quality and best practices
• Generative AI technologies (LLMs, vision-language models, speech models)
• Eval frameworks, agent tooling, RAG pipelines, prompt engineering
• Large-scale distributed training of LLMs (multi-node GPU/accelerator clusters) with PyTorch FSDP2, Megatron-LM, NeMo, DeepSpeed, or similar