Key Takeaways

  • Former commodities trader Anekha Sokhal built Moshi by pairing Bloomberg market data depth with Claude text comprehension and a Netflix-style recommendation engine.
  • The Bloomberg Terminal dominates chat liquidity and raw pricing data, but fails to surface indirect cross-market dependencies.
  • Moshi uses an internal connected graph to flag second- and third-order derivative effects, such as an energy equity filing that shifts crude oil pricing.
  • AI Engineer NYC curator swyx runs a dedicated LLM RecSys track every year because engineers frequently ignore recommendation systems in favor of chat boxes.

Traders Miss the Real Threat Without Connected Graphs

When an oil trader sits at a desk, their terminal displays crude pricing, order flow, and inventory statistics. What they miss are the events taking place two steps removed.

Sokhal points directly to this blind spot. During commodity trading, price shocks rarely originate only within the target market. A sudden announcement by an energy equities company might alter pipeline capacity or capital expenditure, shifting crude balances days later. An oil trader cannot read every corporate filing across the energy sector. As Sokhal puts it: “If you're an oil trader and there's an energy equities company that's doing something which would indirectly impact oil, you need to know that right now.”

Existing terminals do not automate these links. Bloomberg won market share through data depth and chat liquidity, the messaging channels where brokers execute trades. Yet Bloomberg keeps assets in isolated buckets, forcing humans to construct causal connections in their heads. Sokhal argues that manual tracking fails modern markets: “You will get that from a connected graph which is showing that to you because if those two information points are coming out right now the only way you can do it is either a human which is being able to connect those or by a mimicking human sort of like a recommendation system which can do that.”

Search Interfaces Fail When Users Do Not Know What to Ask

Most AI startups build an empty text box. They hook up an LLM, prompt the user for input, and assume the customer will run the right query. In commodities trading, that assumption collapses.

Traders do not query events they do not know exist. A crude trader will never type the ticker of an obscure equity stock into a prompt. If intelligence depends on manual search, it arrives too late. Sokhal bypassed the standard conversational model to build a hybrid: “I wouldn't say we're a new Bloomberg. I'd say we're a different type of software which is not, I think this is the easiest way to say it and I don't know how controversial this is, but if Bloomberg and Claude and Netflix had a baby it would be Moshi.”

The recommendation engine is what enterprise AI usually misses. Netflix does not wait for a search query; it surfaces media based on user behavior. Moshi applies that mechanism to portfolios by using a connected graph. Sokhal explains: “We actually have a recommendation system which we've internally built which is if you're a trader and you're looking at one thing, you actually care about the second and third order derivatives right now which you wouldn't immediately know.”

This aligns with what swyx observes across the industry. At AI Engineer NYC, swyx hosts a track dedicated to recommendation systems. “I think RecSys is one of the most important parts of AI,” swyx said, noting that “as a curator of AIE, I actually have every year an LLM RecSys track and engineers often don't appreciate how important it is.” Pure chat interfaces place the cognitive work on the user. Recommendation engines eliminate it.

What to Do With This

Audit your product interface this week. If your primary workflow requires the user to type a query into a search box, design a prototype that pushes answers proactively. Map your primary user persona to the adjacent datasets they never check, and configure a rule that flags cross-market changes directly on their dashboard.