The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi

Summary · 3 min read

Sokhal studied mathematics, operational research, statistics and economics at Warwick, then traded emerging markets and commodities at JPMorgan, with most of her trading career in energy derivatives. She came to the US on a Fulbright scholarship, did machine-learning research published at ICLR with Stability AI and collaborated with NASA before founding Moshi.

Sokhal says Moshi connects market data with traders’ private sources to surface timely, portfolio-relevant recommendations. She argues that recommendations and interface design work together: software needs to push useful information into a user’s workflow, not just offer a dashboard. She describes filtering noisy sources, showing context when reports conflict, and limiting how customer data is shared with third parties. Sokhal and Shawn Wang also discuss pressure on AI engineers to build systems alone, conference networking and Moshi’s priority of finding customers over hiring.

Minute by minute

1. 0:00

   AI engineers face one-person pressure. Shawn Wang says engineers at the conference feel pressured by management to build entire systems themselves, even though those systems would typically take 10 people. He says many feel overwhelmed by that expectation.

2. 0:53

   Sokhal moved from trading to AI. Anekha Sokhal says she studied mathematics, operational research, statistics and economics at Warwick, then joined JPMorgan’s emerging-markets desk and moved into commodities and energy derivatives. She left trading to pursue a data company, came to the US on a Fulbright scholarship, and later did machine-learning research published at ICLR with Stability AI and work with NASA.

3. 5:39

   Moshi turns market data into insight. Sokhal says Moshi provides real-time insights for financial traders, starting with commodity, energy and macro markets. She says it processes 300 new things in the time it takes her to say a sentence—about four seconds—and aims to deliver answers in 30 seconds that might otherwise take days.

4. 6:46

   Moshi combines public and private data. Sokhal says Moshi looks at futures prices and technical indicators, supply-and-demand fundamentals, news and alternative data such as podcast audio and Twitter. It also incorporates users’ private sources, including subscriptions and internal company data, then cleans and transforms the inputs into a connected graph.

5. 8:56

   Recommendations connect market signals. Sokhal says Moshi is not a new Bloomberg, describing it instead as Bloomberg, Claude and Netflix combined. Its recommendation system connects related signals—for example, an energy-equities event that could affect oil—so a trader sees implications that a news stream may miss.

6. 12:08

   Software has to fit workflows. Sokhal says users can build their own dashboards, but getting someone else to use one becomes a user-interface problem. She argues that platforms can draw users in by pushing relevant data to them, and that Moshi sees interface design and recommendations as linked.

7. 14:36

   Trusted sources need context. Sokhal says commodity traders often rely on trusted voices, including suppliers and LNG facilities that post updates. She says Moshi filters sources and shows a confidence score with the surrounding conversation so users can assess a claim in context.

8. 16:05

   Disagreement can be a signal. Sokhal says reports from Reuters, Bloomberg and the Wall Street Journal may not agree about activity such as ship movements. She argues that the disagreement itself can be useful data because it may reveal an opportunity where the answer is uncertain.

9. 16:32

   Private data shapes model choices. Sokhal says users can upload messy PDFs to Moshi’s cloud version, where a vision-language model breaks down the documents and adds their contents to the database. She says the company does not want to send customer data to third parties, and that building models depends on a team’s engineering capacity.

10. 18:16

    Conference talks surface demand. Sokhal says traders, consultants and AI engineers approached her at the conference. Some engineers felt pressured to build systems that would usually take 10 people, while others described building tools that users said they needed but then ignored in email.

11. 20:29

    Batch meetings could ease networking. Wang questions whether one-to-one meetings are an efficient use of time. Sokhal says Moshi starts with meetings to understand a problem and is planning a webinar to manage inbound; they discuss roundtables and reserved conference rooms as ways to handle repeated conversations.

12. 23:34

    Moshi prioritizes customer meetings. Sokhal says meeting potential clients is a higher priority than hiring, even though Moshi plans to hire. One job listing drew 2,500 applicants in four days, and she says the challenge is finding people who both have the problem and want to buy.

13. 25:45

    Wang moved from trading into AI. Wang says he left hedge funds for tech, then moved quickly into AI as the field emerged. He started an AI engineering post, the conference and media work; he says he misses trading but finds his current work fulfilling because it lets him meet people.

Published Oct 20, 2026.