Key Moments
The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi
Latent Space PodcastKey Moments
AI can now process data at speeds 100,000x faster than humans, but building trust in its recommendations remains a major hurdle for traders.
Key Insights
Moshi processes data 100,000 times faster than a human can speak, analyzing 300 new data points every 4 seconds.
The platform integrates structured data (market time series, technical indicators) and unstructured data (news, X/Twitter, podcasts, audio) with private user data.
Moshi's recommendation system aims to replicate human intuition by connecting seemingly unrelated data points, akin to Netflix's recommendation engine.
The cost of raw, unstructured data like the Twitter (X) Firehose can be as high as $300,000 per month for hedge funds.
AI engineers face pressure to single-handedly build complex systems that would typically require teams of 10 people.
The AI Engineer conference provided significant business value, with Moshi extending their stay to accommodate numerous meeting requests from traders and AI engineers.
Bridging the gap between trading expertise and AI innovation
Anekha Sokhal, founder and CEO of Moshi, brings a unique perspective to the AI and finance world, stemming from her background as a former commodity and energy derivatives trader. After earning a degree in Mathematics, Operational Research, Statistics, and Economics (MORSE) from Warwick, Sokhal was drawn to the fast-paced, data-driven nature of trading. Her career at JP Morgan, particularly in energy derivatives, exposed her to the fundamental drivers of commodity markets and the increasing importance of data in decision-making. This experience, coupled with a desire to innovate, led her to entrepreneurship. A Fulbright scholarship brought her to the US, focusing her research on machine learning, which culminated in work with Stability AI and the NASA, paving the way for the creation of Moshi.
Moshi's approach to real-time market intelligence
Moshi aims to provide financial traders with real-time insights by processing an immense volume of data at speeds far exceeding human capacity. Sokhal illustrates this by stating that in the 4 seconds it takes to say a sentence, Moshi has learned 300 new pieces of information. The platform categorizes data into structured and unstructured types. Structured data includes market time series, technical indicators, and fundamental supply/demand metrics. Unstructured data encompasses news, social media (like X/Twitter), and even audio from podcasts. Crucially, Moshi also integrates private user data, such as analyst recommendations, subscriptions, and proprietary internal data (e.g., LNG ship movements), creating a comprehensive data ontology built on statistics and machine learning. This n-dimensional graph representation allows Moshi to deliver novel insights to users every time they access the platform.
Beyond Bloomberg: A Netflix-inspired recommendation engine
While often compared to Bloomberg, Moshi distinguishes itself by functioning more like a 'Netflix for traders.' Sokhal emphasizes that Bloomberg excels in specific areas, like its chat function, which traders use for high-value transactions. However, Moshi's core innovation lies in its proprietary recommendation system. This system is designed to identify and surface derivative insights—connections between data points that a human might not immediately recognize. For instance, if an energy stock company makes a move that could indirectly impact oil prices, Moshi aims to alert an oil trader to this connection instantly. This is achieved not through a simple news feed, but via a connected graph that visualizes these relationships, mimicking the intuitive, yet complex, recommendation logic found in platforms like Netflix. The goal is to proactively present information that users need, even before they realize it.
Navigating the data deluge: Trust and alternative data costs
The sheer volume of available data, particularly from sources like the X (formerly Twitter) Firehose, presents significant challenges. Sokhal notes that accessing the full X Firehose can cost around $300,000 per month for entities like hedge funds. This raw data is often noisy, with much of it being irrelevant, speculative, or even manipulative. To address this, Moshi filters information by identifying trusted sources within specific industries, such as commodity suppliers or LNG operators, whose public statements can serve as rapid status updates. When presenting information, Moshi assigns a confidence score and highlights discrepancies between different sources (e.g., Reuters, Bloomberg, WSJ) as a valuable data point itself, indicating areas of uncertainty and potential opportunity. This approach helps users discern signal from noise.
Integrating user data and the role of vision-language models
Moshi tackles the challenge of integrating diverse user data, including messy PDF documents. For this, they utilize a vision-language model (VLM) to parse and incorporate this information into their database. The platform also connects directly to users' systems where feasible. The decision to build or outsource VLM capabilities is strategic; while third-party providers exist, Moshi prefers to process sensitive client data in-house to ensure security and control over how the data is handled. Having a dedicated engineering team, many with experience from NASA, further supports their desire to develop and own their models, aligning with their overall mission to create bespoke AI solutions for finance.
AI engineers under pressure and the value of conferences
A recurring theme at the AI Engineer NYC conference, as highlighted by Sokhal, is the immense pressure on AI engineers. Many feel overwhelmed by management's expectation that they act as entire teams, tasked with building complex systems that would historically require ten engineers. This is compounded by the difficulty of ensuring their creations are actually adopted into user workflows, turning a technical problem into a UI and adoption challenge. Conferences like AI Engineer NYC are crucial, not just for networking but for validating product-market fit and identifying potential partnerships. Sokhal's experience at the conference led to an unexpected extension of her stay and a packed schedule of meetings, underscoring the strong demand for AI solutions in finance.
From trader to founder: The journey and future of Moshi
Sokhal's transition from a high-stakes trading career to founding Moshi reflects a broader trend of leveraging deep domain expertise with cutting-edge AI. The overwhelming positive reception at the AI Engineer conference, with numerous traders, fund managers, and AI engineers seeking follow-up meetings, demonstrates a clear market need for Moshi's offerings. The demand for Moshi's insights is so high that they are planning to host webinars to handle the volume of inquiries more efficiently than one-on-one meetings. This strategic move acknowledges the scalability challenges inherent in direct client engagement and aims to broaden their reach while continuing to prioritize client acquisition and development.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●People Referenced
Common Questions
Moshi is a software company that provides financial traders, particularly in commodity and energy markets, with real-time insights. It processes massive amounts of data to identify key information relevant to a trader's portfolio.
Topics
Mentioned in this video
A financial data and software company whose chat feature is noted for its significance in trading, though Moshi differentiates itself from being a direct competitor.
A known vendor for document processing, mentioned in the context of whether Moshi builds its own models or outsources.
Mentioned as a hypothetical parent alongside Bloomberg and Claude for Moshi, highlighting its recommendation system capabilities.
A platform where Annika receives a high volume of job applications and also where people reach out for meetings after her talks.
Annika Sokhal's former employer where she worked in emerging markets and commodity derivatives trading.
A hedge fund where the host previously worked and was one of the first to implement the Twitter/X Firehose.
An organization Annika published with during her machine learning research in the US.
More from Latent Space
View all 266 summaries
76 minHigh Taste Testers: OpenAI Shipped Everything. What Actually Works?
110 minFrom Wet Lab to RL Env: 10,000 Drug-Discovery Campaigns in a Weekend — Ian Quigley, Leash Bio
82 minRobots That Predict the Future: Dyna-2 & Physical AI — Jason Ma, Dyna Robotics
29 minHow to kill your SaaS: what we learned replacing a $43K/yr tool with AI agents
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free