Tech Stack

Description

You will drive the architectural decisions behind Miro's Intelligent Canvas, researching and shipping novel AI architectures that combine LLMs, computer vision, and graph neural networks. Your work will bridge academic theory and production-grade systems, tackling ambiguous problems like generating UML diagrams from rough sketches.

Requirements

• PhD or equivalent deep industrial experience in CS, Math, or Physics
• Track record of patents, impactful open-source, or first-author publications in top-tier conferences
• Expertise in PyTorch or JAX, including complex loss functions and distributed training
• Deep experience in Generative AI, Graph Neural Networks, or Geometric Deep Learning
• Clean production code with understanding of accuracy-latency trade-offs

Responsibilities

• Design and train custom multimodal architectures fusing text, sketches, diagrams, and code
• Bridge academic research (NeurIPS, ICML, CVPR) into scalable low-latency production systems
• Define technical strategy on model selection, build vs. buy, and fine-tuning approaches
• Mentor research engineers through code reviews, paper reading groups, and experimental design
• Solve ambiguous problems with no existing solutions, such as generating valid UML from sticky notes