Machine Learning Engineer | Creative Ads Optimization

The Opportunity Luma AI is a full-stack AI lab building Multimodal AGI. To truly understand the world, models must learn from audio, video, and images. We train our own foundational models and build the products that utilize them. You will join a confidential, founding engineering team tasked with building a new category of creative tools from the ground up. We combine the massive compute resources and capital of a major tech company with the agility and individual impact of a focused startup.

Where You Come In You will operate with high agency as a founding engineer, defining the technical architecture for a new generative product focused on ad creation. This is a role for a builder who thrives in ambiguity and prefers inventing solutions over maintaining existing platforms. You will serve as the bridge between our research team and the product, translating raw model capabilities into a powerful engine that optimizes creative work.

What You Will Build The Optimization Engine: Architect the core ML systems that translate the nuances of ad effectiveness into quantitative signals, guiding our AI models to generate high-performance content. Novel Capabilities: Collaborate with researchers to apply state-of-the-art techniques, including efficient finetuning of vision-language models, to solve specific creative challenges. 0-to-1 Architecture: Design and build the initial backend infrastructure that powers our creative platform, ensuring it can scale from a prototype to a global product.

The Profile We Are Looking For Founding Mindset: You have a history of creating, launching, and building new systems from scratch and thrive in environments where you define the spec as you build. Domain Expertise: You possess a deep, quantitative understanding of the ads ecosystem, specifically regarding creative optimization, personalization, or relevance. Technical Fluency: You have expert-level command of Python and hands-on experience finetuning or training generative models.