Sooth Labs

Sooth Labs (Sooth) is a Pittsburgh-based AI company developing what it calls a global foresight engine: a continuously trained world model intended to produce calibrated probability estimates for long-horizon, cross-domain questions. The company frames existing forecasting as fragmented — prediction markets cover discrete events, time-series models project trends, financial markets price risk, and climate models simulate physical systems — with no single system reasoning about how those domains interact. It argues that large language models, optimized for plausible next-token prediction, do not maintain a persistent world model, learn continuously from outcomes, or generate properly scored probabilities.

The product is described in three components. The Sooth Engine is a continuously trained world model that integrates economic, geopolitical, and operational signals into a unified representation, showing how risks propagate across markets, supply chains, and regions. Sooth Chronology uses autonomous agents to collect and prioritize signals relevant to a customer's decisions, capturing events as they occur and identifying coverage gaps. Sooth Proof cryptographically commits each forecast before outcomes are known, creating an auditable performance record; the company states its probabilities are calibrated such that events assigned a 70% probability occur roughly 70% of the time.

Example questions shown on the company's site span grid capacity constraints on U.S. AI infrastructure, Iranian nuclear escalation, deposit beta through an easing cycle, insured hurricane losses, OPEC+ production discipline, reserve-currency composition, GLP-1 drug revenue, Panama Canal reliability, U.S. office vacancy, Taiwan Strait escalation, Eurozone core inflation, voluntary carbon market regulation, semiconductor supply concentration, LBO returns, enterprise pipeline conversion, yen carry-trade unwind, and copper prices.

Founding story

The company was co-founded by Yaser Sheikh, Russ Salakhutdinov, and Chuck Hoover, a team drawn from large-scale industrial AI research. Sheikh founded and scaled Meta's Pittsburgh AI organization over roughly a decade, moving it from pure research into shipped products including Codec Avatars, and is a Carnegie Mellon Robotics Institute faculty member and author of OpenPose and Panoptic Studio. Salakhutdinov completed his PhD under Geoffrey Hinton, was Apple's first Director of AI Research, later served as VP of Research in Generative AI at Meta, and is among the most-cited active machine learning researchers. Hoover directed billion-dollar innovation programs at Meta and leads product strategy and execution at Sooth. An investor account of the founding rationale cites decades of foundational infrastructure work by the team plus newly available compute and data as the reasons a continuously trained world model is now viable.

Business model

The company markets an enterprise offering, inviting prospective institutional customers to request a briefing rather than self-serve signup.

Traction

The company has closed a seed round backed by institutional venture investors and prominent AI research angels, and has assembled a founding team from Meta, Apple, and Carnegie Mellon. Public product metrics have not been disclosed.

Latest developments

In June 2026, Ardent Venture Partners publicly disclosed its participation in Sooth Labs' seed round alongside Felicis, with Yann LeCun and Jeff Dean investing as angels and Meta CTO Andrew Bosworth serving as an advisor.