Tickets and applications are now live for AIE CODE — the last flagship AIE event of the year. At AI Engineer New York, Anekha Sokhal, CEO of Moshi, explained how her background in mathematics and energy derivatives and work at JPMorgan led her to build a tool that connects market information to what matters in a trader’s book.
What Moshi aims to do
Sokhal left trading because she saw a gap: traders must track far more information than a person can read. Moshi’s product is designed to surface real‑time, portfolio‑specific insights for financial traders, starting with commodities, energy markets and macro. She described Moshi metaphorically as what might happen if Bloomberg, Claude and Netflix were combined: market data, an interface to work with it, and personalized recommendations.
Data types and scale
To convey the data throughput, Sokhal said that while she can say a sentence in four seconds, Moshi can learn roughly 300 new things in that same time. The company ingests multiple classes of data:
- structured time series and technical indicators (futures curves, moving averages, statistical indicators),
- fundamentals time series especially relevant in commodity markets (supply and demand signals), and
- unstructured sources such as news, alternative datasets (e.g., X/Twitter, podcasts, audio) as well as the user’s private data (subscriptions, internal signals, proprietary observations).
Moshi cleans and transforms these inputs, applies statistical and machine‑learning tagging, builds an ontology and places the information into an n‑dimensional graph. The client software then pushes contextually relevant items to the user so that every time a trader opens Moshi they see new items tied to their portfolio.
Recommendation system, workflows and UI
Moshi’s internal recommendation system is designed to surface not only directly related items but second‑ and third‑order effects: for example, if an energy equities announcement would indirectly affect an oil position, Moshi aims to flag that connection via its graph. Sokhal emphasized that recommendations and UI are intertwined — pushing relevant data and embedding it into user workflows increases adoption and makes the tool actionable rather than merely informative.
Trusted sources and conflicting signals
Users often follow specific trusted voices in their market verticals. Moshi provides confidence scores and context: if multiple established outlets (for instance, Reuters, Bloomberg, Wall Street Journal) report divergent information, the disagreement itself becomes a signal that can indicate opportunity or uncertainty. The platform therefore surfaces not only claims but also the surrounding evidence and consensus levels.
Document extraction and build vs. buy for models
Moshi has built vision‑language capabilities to extract structured information from messy PDF documents a customer may upload to their cloud instance. Sokhal noted the tradeoff many companies face: using third‑party processors can be expedient, but when customers hand over sensitive data they often prefer that it not be sent to external vendors without clear agreements. Where a firm has engineering capacity, there is incentive to build bespoke models; where it does not, vendor solutions can be appropriate.
The conversation referenced vendors such as Reducto as examples of external document processing tools, but Sokhal stressed that privacy and contractual clarity determine whether data should be routed to third parties.
Reception at AI Engineer New York and customer outreach
Sokhal reported strong inbound interest after her talk: meetings with traders, consultants and AI engineers followed. Common themes included frustration among AI engineers who are told they ‘are the team’ and therefore expected to build systems that would historically require larger teams. Moshi has been fielding many one‑on‑one meetings; to scale outreach the company plans webinars and other formats to triage interest.
On hiring, Moshi has seen heavy inbound traffic: one open role attracted about 2,500 applicants in a few days, creating recruitment noise. Consequently the immediate priority remains meeting clients and validating product‑market fit over rapid hiring.
Career background that led to Moshi
Sokhal studied mathematics, operational research, statistics and economics at the University of Warwick, then entered trading with internships and roles at JPMorgan on emerging markets and commodity/energy derivatives desks. She obtained a Fulbright scholarship to come to the United States, worked on machine learning research (publishing at ICLR in collaboration with Stability AI and engaging on projects with NASA), and these experiences contributed to founding Moshi.
Conclusion
Moshi attempts to reduce the cognitive load on traders by aggregating and connecting vast amounts of structured and unstructured data, combining public and private signals, applying machine learning and graph‑based models, and delivering recommendation‑driven outputs in real time. Sokhal highlighted the importance of handling sensitive client data carefully, deciding when to build versus use vendor models, and prioritizing user conversations to shape product development.
Viewing the full episode
The interview with Anekha Sokhal is available to watch on YouTube and through AI Engineer New York materials; an automated transcript and show notes (including references to Moshi, the GIQ research paper, Reducto and timestamps from 0:00 to 25:55) accompany the episode.
