Vision-Language AI Scientist: Invent Multimodal Systems

Arcade; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Benefits offered by this job

Competitive compensation

Daily catered lunch

Company events

Job summary

Arcade is seeking a Vision-Language Model expert to lead our efforts in teaching foundation models how to evaluate consumer products like an expert appraiser.

You will invent new technologies, design novel architectures, and author proprietary training paradigms when open-source or commercial models fall short, building systems that assess abstract attributes of product images and generate rich captions, price estimates, and aesthetic quality scores.

Qualifications

• Ph.D. in CS, AI, ML, CV, or related field.
• First-author publications in top AI/CV/NLP venues.
• Deep understanding of Vision-Language Models and transformers.
• Expert Python and PyTorch; optimize training loops.
• Industry research and production deployment experience a plus.
• E-commerce/product ML exposure is a bonus.

Responsibilities

• Pioneer novel neural architectures and multimodal integration techniques.
• Lead fine-tuning and adaptation of VLMs for complex CV tasks.
• Develop methods to extract nuanced signals from product images.
• Guide creation and augmentation of specialized multimodal datasets.
• Invent evaluation frameworks and metrics for abstract tasks.
• Collaborate with engineering teams to deploy research into production.

Skills

Vision-Language Models

Multimodal architectures

PyTorch

Python

Deep learning

Research & publications

Education

Ph.D. in CS/AI/ML

Job description

Arcade is seeking a Vision-Language Model expert to lead our efforts in teaching foundation models how to evaluate consumer products like an expert appraiser.

You will invent new technologies, design novel architectures, and author proprietary training paradigms when open-source or commercial models fall short, building systems that assess abstract attributes of product images and generate rich captions, price estimates, and aesthetic quality scores.