This startup says it is not a new Bloomberg, but a recommendation system for traders
The founder positions the company as a combination of Bloomberg, Claude, and Netflix: value lies not in giving traders more data, but in proactively surfacing second-order signals they wouldn't think to look for, like Netflix does.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
In four seconds, Moshi learns 300 new data points
The founder uses an analogy to illustrate the data density Moshi handles: this four-second sentence, she says, coincides with Moshi learning 300 new data points. Moshi is not another chatbot, but a real-time signal provider for commodity, energy, and macro traders—its goal is to deliver in 30 seconds a judgment that would otherwise take days to assemble, shifting the question of ‘which information matters’ from traders' heads into the system.
— AnikaNot a new Bloomberg—a Netflix for traders
When asked if it is ‘a new Bloomberg,’ the founder clearly denies it, saying if Bloomberg, Claude, and Netflix had a child, that would be Moshi. The key difference lies in a homegrown recommendation system: when a trader is focused on one thing, the system knows they should care about second- and third-order relationships—for instance, how an energy company's moves indirectly affect oil prices. These signals won't appear in a news feed; they require an interconnected relationship graph to actively surface them.
— AnikaRecommendation systems go ignored, yet are indispensable
The host notes that he hosts a dedicated recommendation-systems track at AI conferences each year, staffed by engineers who built YouTube, Netflix, Instagram, and TikTok's recommendation algorithms—yet the track is regularly empty. His explanation: recommendation systems are made to ‘look easy.’ Both agree further: anyone can now build their own Claude dashboard, but getting others to actually use it is a UI problem. Solve the recommendation system well, and the UI problem dissolves.
Twitter's Firehose costs $300,000 monthly and remains all noise
The host recalls that during his time at hedge fund Balyasny, it was among the earliest institutions to purchase X (Twitter) Firehose data, at roughly $300,000 monthly, yet data quality was poor—many post nonsense, and others deliberately manipulate markets. This exposes buyers' core concern: after paying for alternative data, they still may not dare trade on it directly.
Disagreement between Bloomberg, Reuters, and the Journal is itself a signal
Moshi's approach to noise is not to filter out uncertain information, but to display it with a confidence score. The founder illustrates: if Reuters, Bloomberg, and the Wall Street Journal disagree on how many ships or planes passed a location, that disagreement itself is a valuable data point—because nobody knows the truth, which precisely signals a potential trading opportunity.
— AnikaCustomer data stays in-house; self-built models follow
On whether to build a custom vision-language model to parse PDFs, the founder names two criteria: first, customer data doesn't go to third parties unless you can verify how they'll use it; second, if the team has engineering capacity—most Moshi staff have NASA backgrounds—building in-house is the better choice. But she acknowledges not every company has that option.
— Anika‘You are the team’: engineers forced to do ten jobs solo
Among those who found the founder at the conference, beyond traders were many AI engineers, all describing pressure from management: ‘We don't need a team—you are the team.’ A system meant for ten people now falls to one. Worse, some build it only to find colleagues say they need it but never actually open it. The problem becomes non-adoption, not non-delivery.
— AnikaOne open role, four days, 2500 applications
On hiring versus sales priorities, the founder says finding customers is now the bottleneck, not hiring—one job posting draws 2500 applications in four days, forcing it to close. As a first-time founder, she admits not knowing at first how much noise saturates the pipeline: applicants using unauthorized work eligibility, for instance.
— AnikaIn their own words · checked verbatim
So a fun way I explain it to people is it takes me four seconds to say this sentence to you, and in the same amount of time, Moshi has learnt 300 new things.
Anika6:35
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.
Anika9:52
I think, like, REX/X is one of the most important parts of AI, and as a curator of AI years, I actually have every year an LLM REX/X track, and engineers often don't appreciate how important it is.
that's like something like a stupidly expensive, like, $300,000 a month, like, uh, data thing that you have to pay for to get all tweets
But the fact that they don't agree is a point of data because is there an opportunity there? Because no one knows.
Anika16:12
Because people now say, "We don't need a team, you are the team." And so it's like, hey, I need to now build this whole system myself, but that system would be typically done by 10 people.
Anika19:28
We had one job up, we had to close it. It was open for four days. We had 2,500 applicants.
Anika24:54
Figures
| Moshi's processing speed | 300 new data points learned in 4 seconds | 6:35 |
| Single job opening | 2500 applications in 4 days | 24:54 |
| UK Fulbright Data Stream spots | 1 for all of the UK | 4:14 |
Glossary
- RecSys / Recommendation system
- Algorithms that match users to content, powering Netflix, TikTok, and similar platforms.
- MORSE / Mathematics / Operations / Statistics / Economics
- Warwick University's interdisciplinary degree combining mathematics, operations research, statistics, and economics.
- Fulbright Scholarship
- US government–funded international exchange scholarship program.
- ICLR / International Conference on Learning Representations
- A leading machine-learning conference.
- Vision Language Model
- An AI model that understands both images and text, used to parse documents like PDFs.
How to listen
Commodity and energy traders, buy-side alternative-data procurement, or AI engineers being asked to be one person doing one team's work.
22:00–23:30, conversation about conference organization logistics and personal grooming; unrelated to the product itself.