Staff ML Research Engineer, Multimodal Structure & Marengo

Seoul, South Korea (Hybrid) · Engineering · AI

Company: Video models turn footage into data that people and agents can find and use.
Role: Own Marengo models, training, structure and production APIs.

What you will do

We rewrote the responsibilities and qualifications in plain language instead of copying the original posting.

Set multimodal structure, reusable and addressable units, and embedding-model direction, architecture, training and data strategy.
Own distributed training, GPU and data systems, evaluation and production APIs from packaging and design through optimized inference and reliable operation.
Connect structure, semantics and retrieval; align Agent, Search, Product and Infrastructure and strengthen design reviews and mentorship.

At least seven industry years in computer vision, video understanding or multimodal learning.
Ambiguous research problems driven to delivery and principled tradeoffs under changing constraints.
Deep distributed-training expertise such as kernel optimization or FSDP and production model APIs.
Depth in video, multimodal representations or foundation models and end-to-end structure, embeddings and retrieval systems.
Strong Python/PyTorch and ownership through design, build, deployment and iteration, with publications paired with shipped products.

Billion-parameter training and training operations, reliability, monitoring, fault tolerance and cost optimization.
Large-scale data curation and quality and temporal, spatial or hierarchical multimodal structure.
Temporal video understanding, training/inference throughput, latency and GPU efficiency and architectural technical leadership.

Read the employer’s original job description

A former Korea GM at a global company reads it and tells you what to improve for this role. It uses 1 credit.

Checked 2026-10-17

TwelveLabsStaff ML Research Engineer, Multimodal Structure & MarengoStart a mock application