World Models for Enterprise Processes
MIT News AI · rss · 2026-07-15
This MIT News piece introduces Devavrat Shah's research direction: how to make AI more effective in structured data and real-time decision-making scenarios.
Core Concept
- While traditional AI mostly focuses on text and images, a massive amount of critical enterprise data is structured, such as tables and time series.
- His goal is to design methods capable of making sub-second decisions under limited computational resources.
- He likens this approach to graphical models that "reconstruct precise information from sparse signals."
Ikigai and Celonis
- Shah co-founded Ikigai Labs in 2019.
- Based on years of research from his lab, Ikigai built a foundation model for enterprise tabular and time-series data.
- This model was later acquired by Celonis, where Shah now serves as Chief Scientist.
- The goal is to enable enterprises to plug in their own data and processes to make predictions, simulate different strategies, and optimize decisions.
Perspective
- They are focusing on the relatively overlooked domain of "structured/time-domain data."
- The author believes this focus will lead to a narrower but more powerful and highly valuable technological path.
More from Companies & People
- This week’s NYC AI events include Cerebras, Google DeepMind, and a Physical AI hackathon — round · 2026-07-21
- Opendoor CEO shares a raw turnaround playbook in a rare live interview — morganb · 2026-07-21
- A market map tracks outpatient healthcare agentic AI across front, back and mid office — HealthcareAIGuy · 2026-07-21
- FutureCaribbean’s Buildathon offers $50K, H200 compute, and an NYSE pitch — HeyAmit_ · 2026-07-21
- Reliability is a discipline, not a one-time milestone for engineering teams — rseroter · 2026-07-21
- Cloud giants are winning enterprise agentic AI as enterprises buy trust, not features — DavidLinthicum · 2026-07-21