A Data Company in the AI Era: An Interview with Moshi Founder Anekha Sokhal

From Trader to Founder: The Birth of Moshi

Anekha Sokhal spent nearly a decade as an energy derivatives trader at JP Morgan before earning a Fulbright scholarship to pursue a master's in data science at Rice University. Today she is the founder and CEO of Moshi, a Houston-based real-time quantitative platform. Moshi provides real-time analytics for institutional finance traders, monitoring over 2.5 billion data points, with its internal database growing every minute.

"A human takes 4 seconds to read a sentence. In that same time, Moshi has already learned 300 new things," Anekha describes the product's data throughput. When she was a trader, she couldn't even afford to hire an intern, so all data analysis fell on her shoulders. That's why she decided to build the platform she wished she'd had back then. Moshi's core value isn't about providing more data — it's about answering the questions traders struggle with most: What's real? What's relevant to me? What can I trust?

Anekha's background spans statistics, machine learning, and derivatives trading, and she has published papers at conferences including ICLR. While interviewing at hedge funds, she discovered that nearly every one had the same "data synthesis" pain point. Rather than doing the same work over and over for a single fund, she decided to productize the solution.

Energy derivatives trading is one of the most complex domains in financial markets: traders must simultaneously track futures, options, and swap contracts across crude oil, natural gas, electricity, and other commodities, and respond in real time to shifts in geopolitics, weather, and inventory data. The role demands extremely high data throughput and decision speed, which is precisely why Anekha's nearly decade-long experience at JP Morgan gave her a deep understanding of the pain point: "information overload but scarce insight." The Fulbright scholarship is one of the most prestigious international educational exchange programs in the United States, selecting outstanding talent from around the world each year to study in the U.S. Through this program, Anekha entered Rice University to pursue a master's in data science, completing her transition from trade executor to technology builder. Rice University is located in Houston, and its data science program is known for its intersection of statistics and machine learning — laying the foundation for her later publications at top AI conferences like ICLR and her ability to translate academic methods directly into product capabilities.

Entrepreneurship in Houston: A Big Fish in a Big Pond

"Starting up in Houston, you can be a big fish in a small pond." Anekha readily acknowledges that Moshi is one of the companies backed by Texas's largest VC fund and is virtually a household name in the local AI and finance circles. Houston has deep roots in the energy industry, and the client relationships she built as an energy trader translated directly into Moshi's early users.

But she's also clear-eyed about the fact that AI and finance "hype" is concentrated in San Francisco, New York, London, and Singapore. "You have to stay connected to those cities, because that's where the action is." The Houston community is friendly and people are willing to help, but the city is geographically spread out, lacking the "shoulder-to-shoulder" serendipitous encounters of a walkable city.

The Positioning Philosophy of a "Data Company"

Anekha explicitly calls Moshi a "data company" rather than an "AI company." This isn't a word game — it's a declaration of technical architecture. Moshi's foundation is extensive data engineering and machine learning, with generative AI serving only as the topmost interaction layer. "I want to understand data. Generative AI is how we communicate data more humanely."

She offers a vivid example: at 1 a.m., after scrolling Instagram for 40 minutes, she saw 25 AI ads, screenshotted them all, sent them to her growth lead, and said, "We must never advertise on Instagram." When 25 companies use the same talking points to tout "AI for finance," trust becomes a scarce commodity.

"Investors used to require us to have a .ai domain. Now I don't call myself an AI company. If your product has an LLM in it, that's not cool — that's table stakes."

Generative AI refers to technology that automatically generates text, images, code, and other content based on large-scale pre-trained models, with conversational models like ChatGPT and Claude as typical examples. In the fintech context, many companies treat generative AI as the core of their product, but Anekha's judgment is that generative AI is essentially a natural language interaction layer — it excels at translating structured data into human-understandable expressions, but it's not good at guaranteeing data accuracy, real-time performance, and compliance. That's why Moshi chose to make data engineering and machine learning its foundational capabilities, relegating generative AI to the "interface" position. The deeper logic behind this architectural choice is that what financial traders truly need is trustworthy, traceable, low-latency facts — not fluent but potentially hallucinated text. When 25 companies market themselves with the same "AI for finance" messaging, the quality of the underlying data pipeline is the only non-replicable moat.

Easy to Build, Hard to Run

Anekha points to a divide that the industry widely underestimates: building a prototype with AI is extremely easy, but making a system run reliably in production is a completely different matter. You can spend $200 on a Claude subscription and build yourself a beautiful dashboard within a week. But when you want your mom (who doesn't have a Claude account) to use it, the problems begin.

Enterprise-grade deployment requires far more than a prompt: DevOps, error handling, DDoS protection, API key management, database operations — "these are things no LLM will proactively tell you." Even more insidious is prompt fatigue: change one thing, and you inadvertently break 50 others.

She describes this as a painful period that requires a "detox": people first use AI for everything, then they realize — "I actually want a spreadsheet I need to check with my brain. I want to remember someone's phone number. I want to see an imperfect PDF."

DDoS protection (Distributed Denial of Service protection) refers to security measures that use traffic scrubbing, rate limiting, CDN distribution, and other means to prevent malicious traffic floods from overwhelming service availability. For real-time platforms serving institutional traders, a few minutes of downtime can cause actual trading losses, making DDoS protection not optional but foundational infrastructure. API key management involves encrypted storage, rotation, least-privilege access, and audit trails for keys. Once leaked, attackers can impersonate the platform to call downstream data sources or model interfaces, leading to data breaches and runaway costs. The reason Anekha says these engineering details are "things no LLM will proactively tell you" is that large language models' training corpora lean heavily toward algorithms and product demos, while production-environment knowledge about operations, security, and fault tolerance often resides in the tacit experience of senior engineers. From prototype to production, what teams need to fill in are exactly these unsexy but life-or-death engineering capabilities.

Human Judgment: Claude Can't Build Moshi

Anekha ran a hiring experiment: she used Claude to screen and rank resumes, then brought every candidate into the office for three to four hours of face-to-face interaction, meeting team members and investors. The result: all human interviewers produced highly consistent rankings, while the person Claude ranked first didn't even make the humans' top five.

Another experiment was even more striking: 20 people were given a sales lead research task, allowed to use any AI tools, to select 10 target contacts from LinkedIn's millions. 90% of them gave the same 10 companies.

"Everyone can use the same tools, you're still pulling and pushing the same data, using the same models. If my engineer asks Claude a question, no matter how the prompt changes, the answer is basically the same."

Anekha believes humans' unique value lies in second- and third-order associations and "seeing around corners." Claude excels at summarizing and explaining, but not at intuitive judgment and interpersonal interaction. She mentions a "one-person company" that raised $30 million with impressive social media metrics, yet user comments repeatedly say, "I just want to talk to a real person."

Second- and third-order associations refer to humans' ability, when making decisions, to step beyond direct causation and reason through multi-layered chains of "if A happens, then B, and then C." In a trading context, this means not just seeing inventory data rise, but also associating it with transportation bottlenecks, refinery maintenance schedules, changing weather patterns, and the likely behavioral responses of counterparties. Current large language models are essentially predicting the next word based on statistical patterns. They excel at summarizing and explaining within patterns covered by existing corpora, but lack genuine causal reasoning and intuitive leaps when facing rare or unprecedented combinations of circumstances. Anekha's hiring experiment proves exactly this: Claude can rank by resume keywords, but cannot capture the resilience, curiosity, and team chemistry revealed in face-to-face interaction — which are precisely the key variables in a high-frequency trading team's long-term performance.

The Future of Human-AI Collaboration

Moshi's business model confirms this judgment: clients have started sending over their prompts, saying "you run it for us." Enterprise software is shifting from tool delivery to "managed trust" — clients don't want to run it themselves, don't want to manage security, don't want to remind Claude on a trading day that today is an FOMC meeting.

"Claude can do 90%, but that 10% is exactly where you don't want Claude. Because at the end of the day, when something goes wrong, you want to talk to a person and say, 'Hey, let's figure this out together' — not converse with a chat model."

Anekha's conclusion is pragmatic and measured: AI is a tool for specific scenarios, not a master key that replaces everything. Moshi provides data and confidence, letting traders see blind spots they couldn't see before. But ultimately, the person who presses the trade button and goes to sleep bearing the risk is still human.

One-line Summary

Moshi founder Anekha believes AI is a tool, not a master key — human judgment remains the ultimate moat in financial trading.

Paragraph Summary

Moshi founder Anekha Sokhal transitioned from a JP Morgan energy derivatives trader to a data scientist, building a real-time quantitative platform in Houston that monitors over 2.5 billion data points. She explicitly positions Moshi as a "data company" rather than an "AI company," with data engineering and machine learning at its core and generative AI serving only as an interaction layer. Through her hiring experiment and sales lead research experiment, the article reveals: building AI prototypes is easy, but production-grade deployment faces systemic challenges in DevOps, security, and prompt management. Meanwhile, when everyone uses the same AI tools, outputs become highly homogenized, making humans' second-order associations, intuitive judgment, and interpersonal interaction irreplaceable sources of differentiation. Enterprise software is shifting from tool delivery to "managed trust," but ultimately, the person who presses the trade button and bears the risk is still human.