Foundation Model Data Engineer

About the role

Sciforium is building next‑generation multimodal AI models and a high‑efficiency serving platform. The Foundation Model Data Engineer will own the full data lifecycle, from web‑scale crawling to curated human‑aligned datasets that shape model behavior.

Key responsibilities

• Define and create pre‑training datasets mixing web data, code, books, and technical papers.
• Design petabyte‑scale pipelines for cleaning, exact/fuzzy deduplication, and signal extraction.
• Lead development of post‑training datasets for supervised fine‑tuning, multi‑turn dialogues, and RLHF/DPO preference modeling.
• Drive acquisition and processing of vision and video data for multimodal alignment.
• Build high‑throughput Python scripts using multiprocessing and multithreading to ingest and transform massive data streams.
• Perform statistical analysis on corpora to detect biases, gaps, and quality regressions.

Required profile

• 5+ years of experience building and managing datasets for foundation models.
• Hands‑on experience with petabyte‑scale data that has been used to train production LLMs or large vision models.
• Proven ability to reconstruct massive training sets from raw web crawls and domain‑specific sources.
• Experience creating RLHF/DPO and multi‑turn instruction datasets, including human‑labeling workflows.
• Mastery of data‑at‑scale frameworks such as Spark, Ray, WebDataset, or Parquet.

Required skills

• Python (high‑performance code, multiprocessing, multithreading, memory management).
• Data‑cleaning, deduplication, and large‑scale ingestion pipelines.
• Spark, Ray, WebDataset, Parquet.
• Petabyte‑scale data processing and storage.
• RLHF/DPO dataset creation and human‑labeling workflow management.

What we offer

• Medical, dental, and vision insurance.
• 401(k) retirement plan.
• Daily lunch, snacks, and beverages.
• Flexible time off.
• Competitive salary and equity.