Member of Technical Staff, Multimodal Pretraining Data
Core
Develop and curate multimodal datasets (text, code, image, video) to pretrain AI models for language and visual understanding.
Role type
Senior IC multimodal data engineering and research engineer
Builds
Paired and interleaved datasets (image-text, video-text) and training mixtures for multimodal AI models
Domain
Artificial Intelligence / Multimodal Machine Learning
Deliverable
production ML models
Required skills
Python engineering, distributed processing frameworks (Spark, Ray, Beam), data curation, controlled experiments, quality classification, synthetic supervision, data mixture design, corpus quality measurement
Preferred skills
image-text alignment, interleaved document construction, video frame sampling, synthetic captioning, embedding-based data selection, data mixture scaling, tokenizer design, multimodal benchmark decontamination
Responsibilities
Develop methods to select, filter, deduplicate, and prepare text, code, image, and video datasets at scale; Build paired and interleaved datasets preserving document structure and visual-text alignment; Build quality classifiers and sampling methods and measure impact via controlled experiments; Develop synthetic captioning and supervision methods for image and video data; Design mixtures and curricula across sources, domains, and modalities; Measure corpus quality, check for benchmark contamination, and track source provenance; Work with infrastructure engineers to scale data processing
Seniority
Senior, hands-on IC