LLM Dataset Engineer
Core
Designing taxonomies, filtering heuristics, and post-training pipelines for massive datasets that power foundation LLMs and multimodal models.
Role type
Senior IC LLM Dataset Engineer
Builds
Pre-training corpora, SFT/RLHF/DPO alignment datasets, and multimodal vision/video data pipelines
Domain
AI Infrastructure / Large Language Models / Multimodal AI
Deliverable
production ML models
Required skills
Python (high-performance, multiprocessing, multithreading), petabyte-scale data processing, dataset reconstruction from web crawls, RLHF/DPO pipeline design, data profiling and bias analysis, synthetic data generation
Preferred skills
Computer vision curation, multimodal crawling, taxonomy design for reasoning/coding/math, research background in data-centric AI
Technologies
Spark, Ray, WebDataset, Parquet
Responsibilities
Define mix of web data/code/books/papers for pre-training; implement pipelines for cleaning, deduplication, and signal extraction from petabytes; lead development of SFT and preference modeling datasets; drive acquisition and processing of vision/video data; develop high-throughput ingestion scripts; conduct statistical analysis on training corpora; design pipelines for synthetic data generation
Seniority
Senior, hands-on IC