Deep Learning Engineer (AI/LLM)

Company

hirify.global is a foundational AI lab that develops technologies for scaling AI agents and reducing token cost through model routing, harness optimization, and model training.

What you will do

• Develop, improve, and maintain training frameworks and related infrastructure for applied research projects.
• Optimize training performance and quality through parallelism methods, training SDK improvements, RL environment integration, new model architectures, and training-method research.
• Optimize model serving for high throughput and low latency in production and training environments.
• Participate in critical post-training efforts with applied researchers.
• Adapt code to use advanced hardware accelerators and run training and evaluation experiments for LLM optimization.

Requirements

• 3+ years of software development experience for production systems, with strong Python skills and solid engineering fundamentals including testing, debugging, performance work, and reading unfamiliar codebases.
• Hands-on understanding of end-to-end modern LLM training in PyTorch, including data, model, tensor, and pipeline parallelism; mixed precision; memory-versus-compute trade-offs; and the underlying operation of frameworks such as FSDP, Megatron, or DeepSpeed.
• Deep understanding of LLM architectures, including attention variants, MoE, positional encodings, and KV cache.
• Ability to debug across the stack, including loss spikes, NCCL hangs, log-probability mismatches, and underperforming kernels.
• At least a BSc in computer science, electrical engineering, or a similar field.
• Significant experience in at least one specialization: distributed training; post-training and RL; inference systems; or model efficiency.

Nice to have

• Experience in a distributed, containerized GPU training environment such as Slurm, Kubernetes, or similar, including launching and debugging multi-node jobs and handling images, storage, and scheduling constraints.
• Experience with distributed training stacks such as FSDP, Megatron, DeepSpeed, or JAX-based systems.
• Experience with SFT, DPO, GRPO, online or offline training, rollout-trainer coupling, or frameworks such as verl, slime, prime-RL, TRL, or in-house equivalents.
• Experience with serving frameworks such as vLLM, SGLang, or TensorRT-LLM, including batching, scheduling, KV cache management, and latency or throughput optimization.
• Experience with quantization, distillation, kernel-level optimization, speculative decoding, or model export and serialization for deployment.

Culture & Benefits

• Work alongside applied researchers on foundational AI and large language model systems.
• Contribute to advanced research and engineering across training, inference, post-training, and model optimization.