The Position

OPSWAT is committed to harnessing the power of AI to drive meaningful advancements across all areas of our business. We are currently seeking a highly skilled Senior AI Engineer to lead the deployment and optimization of Large Language Models for efficient, high performance inference on GPU hardware.

In this role, you will own the infrastructure layer that turns our AI models into fast, cost effective, production grade services. You will work closely with the AI/ML Team, the Enterprise & Data Team, and Infrastructure to design local LLM serving environments, maximize GPU utilization, and apply model compression techniques such as quantization from FP32 down to FP8/FP4. This is a chance to work at the deep intersection of models, serving software, and hardware within a global cybersecurity company, delivering the performance that powers OPSWAT's AI products at scale.

What You Will Do

• Local LLM Deployment & Hardware Setup: Design and build local LLM serving environments on GPU hardware, selecting GPU configurations based on VRAM, memory bandwidth, and workload. Install and maintain the full inference stack (NVIDIA drivers, CUDA, cuDNN, inference engines) across single and multi GPU deployments, including tensor and pipeline parallelism.
• Inference Optimization & Model Compression: Optimize LLM models for efficient GPU serving, balancing latency, throughput, memory, and cost. Apply quantization from FP32 to FP16/BF16, FP8, and FP4 using PTQ methods (GPTQ, AWQ, SmoothQuant) and QAT. Design mixed precision strategies and run calibration to minimize quality loss. Apply complementary techniques including pruning, distillation, and sparsity.
• Serving Engine & Performance Engineering: Deploy and tune high throughput serving engines (vLLM, TensorRT-LLM, TGI), optimizing KV cache (PagedAttention), continuous batching, and speculative decoding. Leverage optimized kernels (FlashAttention) and FP8 tensor cores. Build benchmarking for TTFT, inter token latency, throughput, GPU utilization, and cost per million tokens.
• Production Serving Infrastructure: Deploy quantized models with autoscaling, load balancing, and observability. Establish quality regression gates and run A/B tests of quantized vs full precision models on real traffic.
• Research & Propose Innovative Solutions: exploring and implementing novel inference optimization and model compression techniques.

What We Need From You

• Bachelor's degree in Computer Science, Software Engineering, or a related field.
• 5+ years of experience in software engineering, with a focus on ML infrastructure, LLM inference, or model optimization.
• Hands on experience deploying and serving LLMs on GPU hardware in production.
• Strong understanding of quantization and model compression (FP8/FP4/INT8/INT4, GPTQ, AWQ, SmoothQuant, QAT).
• Experience with high throughput inference engines (vLLM, TensorRT-LLM, TGI, llama.cpp).
• Solid understanding of GPU architecture, CUDA, and the memory bandwidth bound nature of LLM inference.
• Familiarity with KV cache, continuous batching, PagedAttention, and speculative decoding.
• Solid understanding of software development methodologies (Agile, Scrum, etc.).
• Good communication, collaboration, and problem solving skills.
• Ability to work independently and as part of a team in a fast paced environment.
• Proficiency in Python; familiarity with C++/CUDA is a strong plus.

It Would Be Nice if You Have

• Experience writing or tuning custom CUDA / Triton kernels.
• Experience with multi GPU and distributed inference.
• Certifications in AWS or Azure architecture.
• Experience with CI/CD and cloud production deployment (Azure, AWS, or GCP), including GPU backed instances.