AI labs that need real non-English or multimodal training data, not synthetic or machine-translated substitutes, are Mundo AI's target buyer. Its four founders, who came out of quant research, Hugging Face pretraining and Binance.US product roles, built the company after hitting the multilingual data shortage directly.
Mundo AI, founded 2024 in Vancouver by four YC W25 founders, builds multilingual and perceptual-intelligence training data. It runs native-speaker data operations in target-language countries, supplying audio, video and text datasets to frontier AI labs building non-English and multimodal models.
Founded: 2024 · HQ: Vancouver, BC, Canada · Team: 10-20 · CEO: Jason Liao · Funding: $500K Seed (Feb 2025), led by Y Combinator with Augment Ventures, Broadshade Investments, GreatPoint Ventures and N1
About Mundo AI
Mundo AI was founded in 2024 in Vancouver, BC by four co-founders who met studying at the University of British Columbia: Jason Liao (CEO), who previously led a quant research team at a C$60B hedge fund and built a fraud-detection model at Tsinghua University; Garreth Lee, who worked on pretraining data at Cohere and was the first Indonesian hire at Hugging Face, where he helped build one of its open pretraining datasets; Naijide Anwaer, a former Platform PM at Binance.US who speaks four languages; and Kenneth Wu, a former quant at a large Canadian pension fund with prior software engineering experience at Amazon Web Services. The founding thesis, per Jason Liao's account, came from firsthand experience doing ML research abroad and discovering how little usable training data exists for non-English languages, a gap the founders argue leaves the roughly 75% of the world that does not speak English underserved by current AI models. Mundo AI went through Y Combinator's Winter 2025 batch and launched publicly in February 2025. The original product, as pitched at YC launch, was a multilingual data library: Mundo AI sets up in-country operations in the places where target-language native speakers live, and uses a proprietary software platform to run data collection, generation, annotation and quality assurance for non-English text and speech datasets, selling into AI research labs building multilingual models. By mid-2026 the company's own site describes a broader scope, positioning Mundo AI as 'the data layer for perceptual intelligence' and saying it works with frontier research teams on data, evaluations and applied research spanning audio, video and other modalities, not text-only multilingual corpora. There is no public pricing, self-serve signup or product demo on the current site; engagement appears to run through direct partnerships with AI labs rather than an off-the-shelf purchase. Mundo AI has raised $500K in a seed round that closed in February 2025, with Y Combinator as the anchor investor alongside participation from Augment Ventures, Broadshade Investments, GreatPoint Ventures and N1. Augment Ventures lists the investment on its own portfolio page, quoting the company's pitch as building 'the infrastructure needed to change' the fact that AI models perform worse in every language other than English. No later funding round or valuation has been disclosed as of mid-2026. The YC company page still lists a team size of 4, matching the founding group, but third-party trackers checked in 2026 put headcount higher: Tracxn recorded 11 employees as of March 2026 and PitchBook recorded 18, while the company has publicly posted at least one Vancouver-based software engineering internship role in 2026. The spread across sources suggests real but modest growth from the founding team rather than a settled headcount, and no figure here should be read as company-confirmed. Mundo AI competes for budget with much larger, better-capitalized data vendors: Scale AI and Surge AI on enterprise-scale data labeling and RLHF pipelines for frontier labs, and Turing on distributed annotator networks. Its stated edge is operational, running native-speaker data collection directly inside the countries where under-represented languages are spoken, a model that is harder for a purely platform-based competitor to replicate quickly, but one that is also capital- and headcount-intensive to scale against incumbents with far deeper funding.
Mission
Building the world's largest and highest-quality multilingual data library so AI models work as well outside English as they do inside it.
Products
- Multilingual Data Library (Training data licensing): https://mundoai.world
- Perceptual Intelligence Data (Audio/video data, evaluations & applied research for frontier labs): https://mundoai.world
Links
Frequently Asked Questions
How much funding has Mundo AI raised?
Mundo AI raised a $500K seed round that closed in February 2025, led by Y Combinator with participation from Augment Ventures, Broadshade Investments, GreatPoint Ventures and N1. No later round or valuation has been disclosed as of mid-2026.
What does Mundo AI actually sell?
Mundo AI licenses multilingual and, more recently, perceptual-intelligence training data (audio, video and text) to AI research labs. It runs data collection, generation, annotation and quality assurance through in-country operations staffed by native speakers of the target language, rather than synthetic generation or machine translation. There is no public pricing or self-serve signup; engagement is through direct partnerships with AI labs.
Does Mundo AI publish compliance certifications or a trust center?
No. Mundo AI has not published SOC 2, ISO 27001, GDPR, or any other compliance certification, and it has no public trust center. It also has not disclosed a policy on provenance, consent, or compensation for the native speakers who contribute source data.
Who competes with Mundo AI?
Mundo AI competes for the same frontier-lab data budgets as much larger, better-funded vendors: Scale AI and Surge AI on enterprise-scale data labeling and RLHF pipelines, and Turing on distributed annotator networks. Mundo AI's edge is running native-speaker operations directly inside under-represented language markets, but it is a small, thinly-funded company against incumbents with hundreds of millions raised.
How do AI labs start working with Mundo AI?
There is no self-serve product or public pricing page; the company describes working directly with frontier research teams on bespoke data, evaluation and applied-research engagements. Interested labs would need to contact Mundo AI directly through its website rather than sign up for a hosted product.