Luma AI

Compensation

Full Description

The Opportunity Luma AI is a full-stack AI lab with a mission to build Multimodal AGI. We believe that true intelligence requires the ability not only to process information, but also to imagine and create it. You will be joining a confidential, founding team with the resources of a major tech giant and the speed of a research skunkworks. We are working through fundamental problems in creative AI: Can a model understand "quality" and "persuasion" the way a human creative director does? Where You Come In We are looking for a builder who can turn subjectivity into math. In this role, you will bridge the gap between our foundational Omni models and the complex requirements of the advertising world. Your challenge is to architect the systems that allow our AI to "see" creative work, evaluate its potential effectiveness, and optimize it iteratively. You aren't just predicting a click; you are engineering the feedback loops that teach a model how to be creative. What You Will Build Aesthetic Evaluation Metrics: Define and implement the "eyes" of the system, quantitative metrics that correlate with human perception of quality, brand alignment, and engagement. The Optimization Loop: Build the backend architecture that uses performance data to finetune Vision-Language Models (VLMs) in real-time, effectively creating a self-improving creative engine. Generative Experiments: Design the technical framework for testing thousands of creative variations against one another, discovering novel patterns in what makes an ad successful. The Profile We Are Looking For The Scientist-Engineer: You are as comfortable reading a research paper on VLMs as you are writing production Python code. You prefer problems that don't yet have a Stack Overflow answer. Domain Translator: You have a deep background in AdTech (creative optimization, ranking, or relevance) and can translate abstract marketing goals (like "brand lift") into concrete loss functions and engineering constraints. Curiosity: You are fascinated by the intersection of art and algorithm.