AI startup specializing in visual reasoning — building multimodal models capable of spatial and physical reasoning without converting images to text.
Elorian AI Inc. is a Palo Alto-based startup focused on multimodal visual reasoning. The company aims to build the foundations of visual reasoning — AI systems that natively understand and interpret the visual environment the way humans do, without translating images into text. The team comprises former researchers from Google Brain, DeepMind, Apple Intelligence, and Salesforce Research, including Andrew Dai (co-creator of Gemini and PaLM 2), Seth Neel (chief architect of Apple's MM1), and Seo-Jin Bang (former Harvard professor). In April 2025, the company raised $55M in pre-seed funding from Striker Ventures, Menlo Ventures, and Altimeter Capital, achieving a valuation of approximately $300M. Elorian's models are designed for engineering, robotics, medicine, life sciences, and precision agriculture — wherever spatial and physical reasoning is critical.
Founders
Former Google Brain & DeepMind researcher. First author of the 2015 paper introducing LM pretraining + supervised fine-tuning. Co-led early MoE models including GLaM. Lead for Gemini data area and co-led PaLM 2 pretraining.
Former Google Research & Apple researcher. Led multimodal research including Apple's first public multimodal foundation model (MM1). Pioneered phrase grounding, UI understanding, and multimodal reasoning.
Former Harvard Professor researching data-centric generative AI (machine unlearning, data attribution, privacy), Google Research/Gemini Data. Co-Founder/CTO of ML energy startup (100+ FTEs). UPenn PhD, Harvard Math.
Formerly Apple (Apple Intelligence Foundation Models, multimodal research, visual grounding, computer-use agents, UI understanding) and Google Research (generative UIs). Berkeley CS PhD.
Former Google Research: Gemini Data and pretraining. Specialist in Data Mixture Optimization, Bilevel Optimization, and Federated Learning.
Former SRI International researcher: multimodal perception and learning, human behavior modeling, temporal causal discovery, and real-time ML. Clinical assessment, driver monitoring, and visual skill understanding.
Former Tech Lead for multimodal research at Salesforce Research. Specialization: multimodal foundation models, multimodal LLMs for understanding, and unified multimodal modeling.
Former Google DeepMind: Gemini Data and Agents. Co-creator of Google's first public agent (Project Mariner). Work on geospatial computer vision and chemical formulation stability modeling.
Former Google DeepMind: Gemini post-training, robust reward self-improvement, prompt and data optimization. Co-creator of OSS Vizier and Optformer. Berkeley PhD.
Former core architect for Gemini at Google DeepMind. Pioneered the inference and distillation strategies that define the state-of-the-art in LLMs.
Classification
External links
