01 / 11

Where great datasets are built, found, and used.

Agents can now build any dataset in an afternoon. Mostly Right is where people and agents find and use the best ones.

THE SHIFT

02 / 11

Building data used to take a team. Now it takes a prompt.

A good dataset used to need a data team and weeks of work. Now an agent does that work in an afternoon. So more datasets get built and shared than ever.

Without agents
With agents

Over 1M public datasets on Hugging Face, roughly doubling every year. Source: Hugging Face Hub, Oct 2026

THE PROBLEM

03 / 11

More datasets than ever, and no place they all belong.

Builders ship them anywhere — Kaggle, Hugging Face, a personal drive — and there’s no common layer on top.

Hard to check

You often can’t see the rows or the source until you download it.

Hard to use

Answering one question can mean pulling down the whole file.

Hard to trust

You can’t tell if it’s good without digging into who else uses it.

On Hugging Face, the top 1% of datasets get 77% of downloads. Most get almost none. Source: Hugging Face Hub stats, Oct 2026

THE SOLUTION

04 / 11

We make great datasets easy to build, find, and use.

Agents get the tools to build good datasets that are documented, searchable, and easy to query. Every dataset gets a public page and an audience of people and agents.

Build

  • Models to clean and classify data
  • Our models predict, fill gaps, and fix errors
  • Tools to refresh, query, and handle any format

Find and use

  • Smart, fast search inside any dataset
  • Query it by API or MCP, no download needed
  • Likes, uses, and views show what’s good

THE FLYWHEEL

05 / 11

Good data brings users. Users bring more good data.

We start by rebuilding the most popular datasets from other platforms, fresher, searchable, and ready to query. Users follow, and publishers and data vendors follow them. Once vendors are here, paid datasets become a natural next step, listed right beside the free ones.

THE TEAM

06 / 11

Built by founders who have shipped, sold and published.

Twenty years ago Vu and Vojta built a mobile game together in high school. An exit and a PhD later, Vu met Rob at Marinade and introduced him to Vojta. The three of us teamed up to start Mostly Right in early 2026.

Vu Hoang Anh

Vu Hoang Anh

CEO

Built 13 dev tools that generated $1M in revenue, enough to bootstrap Avocode. Then took it through 500 Startups to $2M ARR before it was acquired.

Rob Tarabcak

Rob Tarabcak

CTO

Founded Cinnamon and raised a $3.9M seed for it. Then spent two years leading product at Marinade Finance, where he and Vu first worked together.

Vojta

Vojta Havlicek

RESEARCH

Took a masters at Imperial and a quantum computing PhD at Oxford, and is now at IBM Research. His first-author Nature paper has almost 4,000 citations.

Backed by Hustle Fund 2080 Ventures

TRACTION

07 / 11

29,165 views. 3,284 uses.

111 public datasets · 30 Sept 2026

DISCOVER DATASETS

08 / 11

Find data the way you find code.

GitHub made code discoverable and the world built on it. We’re doing the same for datasets. Teams find us through the free catalogue, then explore and use the data through one API.

Explore community ↗

WHY WE BUILT THIS

09 / 11

We started with our own data.

We built a data layer for our own models. Working with other teams showed us how much work it takes to make data usable, and keep it that way.

Built for ourselves first

We built this data layer to trade prediction markets quantitatively, and our own strategies still run on it.

Ten teams, two paying

We opened it as an SDK. Every team told us it solves a real problem, but they need more data sources.

Agents do the data work

Adding sources by hand did not scale. We built agents to clean, join, and maintain the data those teams needed.

THE OPPORTUNITY

10 / 11

Companies already spend billions on data.

Companies pay for data, then pay engineers to clean, integrate, and maintain it. We help teams build and publish datasets that others can discover, explore, and query.

Estimated engineering spend on public data $2–4B

LinkedIn lists about 190,000 data engineers, a Fivetran survey found they spend 44% of their time building and maintaining pipelines, and we assume a third of that is public data.

Our estimate · LinkedIn, 2024 · Fivetran, 2021

THE VISION

11 / 11

Open up the world’s data.

More datasets are being built than ever. We’re making every one of them easy to find, trust and use, so anyone, anywhere, can work with the world’s best data.