Staff Applied Scientist - VLLM Inference

Changing the world through digital experiences is what Adobe’s all about. We give everyone—from emerging artists to global brands—everything they need to build and deliver outstanding digital experiences. We’re passionate about empowering people to develop beautiful and powerful images, videos, and apps, transforming how companies interact with customers across every screen. We’re on a mission to hire the very best and are committed to crafting outstanding employee experiences. Everyone is respected and has access to equal opportunity. We realize new ideas can come from anywhere in the organization, and we know the next big idea could be yours!
Adobe Firefly’s ASML group invites research scientists and engineers passionate about conditional generation and editing of large generative AI models. This role emphasizes images and videos. We strive to advance generative AI technology while guaranteeing models possess excellent quality and control.
As an Applied Scientist, you will define technical strategy for multimodal data intelligence systems, architect and optimize distributed LLM/VLM inference platforms, and develop innovative solutions for automated captioning, tagging, metadata enrichment, and dataset creation. You will work at the intersection of research and engineering, collaborating with teams across modeling, infrastructure, data, evaluation, and product to deliver high-quality AI capabilities at scale.
You will have the opportunity to influence the next generation of Adobe Firefly models by improving data quality, model efficiency, and scalable AI infrastructure used by millions of creators worldwide.
Job Responsibilities

• Architect and optimize distributed multimodal inference pipelines for large-scale image, video, and audio captioning, tagging, and metadata generation.
• Drive LLM/VLM inference optimization, including batching, scheduling, quantization, model serving, caching, and GPU utilization to maximize throughput and cost efficiency.
• Build scalable data generation workflows using state-of-the-art vision-language and multimodal foundation models to improve training data quality.
• Lead technical strategy for automated dataset annotation, filtering, quality scoring, deduplication, and metadata enrichment across multimodal datasets.
• Design distributed processing systems capable of handling billions of media assets across heterogeneous compute environments.
• Collaborate with research teams to productionize new LLM/VLM capabilities while ensuring scalability, reliability, and operational efficiency.
• Partner with infrastructure teams to improve distributed execution frameworks, storage systems, and inference services.
• Drive cross-functional alignment across data, research, infrastructure, evaluation, and product teams on multimodal data processing strategy.
• Mentor engineers in distributed systems, scalable ML infrastructure, and multimodal AI engineering best practices.

• Ph.D. or M.S. in Computer Science, Machine Learning, or a related technical field, with significant industry experience designing and deploying large-scale distributed ML systems.
• Deep expertise in large language models (LLMs), vision-language models (VLMs), or multimodal foundation models, with hands-on experience building, optimizing, and serving inference workloads at scale.
• Strong background in distributed systems, large-scale data processing, and cloud-native ML infrastructure, with experience using frameworks such as Ray, Spark, Dask, Kubernetes, or equivalent technologies.
• Proven experience optimizing large-scale LLM/VLM inference systems, including techniques such as batching, parallelism, quantization, model serving, GPU utilization optimization, and latency/throughput tuning.
• Experience building high-throughput multimodal data pipelines for automated image, video, and audio understanding tasks, including captioning, tagging, OCR, metadata extraction, and semantic indexing.
• Hands-on experience with modern ML inference and serving frameworks such as vLLM, TensorRT-LLM, SGLang, Triton Inference Server, TGI, or equivalent technologies.
• Experience managing and processing petabyte-scale multimodal datasets using distributed storage and data processing systems.
• Familiarity with multimodal embedding models, retrieval-augmented systems, vector search infrastructure, and data quality evaluation methodologies.
• Strong software engineering skills in Python and PyTorch, with a track record of developing reliable, production-grade distributed ML systems.
• Excellent communication and collaboration skills, with the ability to influence technical strategy and drive alignment across research, infrastructure, and product teams.