Moshi CEO: 300 new things learned in 4 seconds, à la Bloomberg, Claude and Netflix

Sokhal's path to Moshi ran through mathematics at Warwick, a JPMorgan internship that became a career on emerging markets and then energy derivatives, and a Fulbright scholarship to the US, where she worked on machine learning research published at ICLR with Stability AI and in collaboration with NASA. That background shapes how she describes the product: a trader breaks the market into structured and unstructured data, from futures time series and technical indicators to fundamentals, news, X posts, podcast audio, and private user data such as analyst contacts and internal ship-tracking counts. Moshi cleans, tags and organizes that into what she calls an n-dimensional graph, then pushes new items to the user's portfolio.

The design reflects her view that dashboards alone do not change behavior. She argued that when someone builds their own Claude dashboard, getting another person to adopt it becomes a UI problem — and that push-based recommendations, the way Spotify or Instagram hook users, are how software actually gets used. Moshi's recommendation system aims to surface indirect links, such as an energy equities company whose moves would indirectly affect oil.

On data handling, Sokhal said Moshi built a vision language model to read messy PDF data because customers uploading proprietary information should not have it sent to a third party; she framed building versus buying models as a question of whether a company has the engineering team. The conference also surfaced a pressure point: AI engineers told her management now says 'you are the team,' asking one person to build what would typically take ten.