New users are blank slates.
Every new customer arrives with no signal, no context. Your AI serves them something generic. The data exists. It’s stranded in platforms they already use.
AI gets better when it understands the human in front of it.
We build the infrastructure that makes that possible.
The model is not the bottleneck.
Getting the model to understand a user's context is.
Training data abundance is not training data quality.
Synthetic data is fast and scalable. It also behaves like synthetic data. Models trained on real, consented human data perform differently.
83% of users will share data for a better experience.
The willingness is there. The infrastructure to capture it, with consent documented at the record level, is not.
Building data integrations one by one doesn’t scale.
Every platform API is different. Every integration breaks. There is a better entry point.
Products
Infrastructure for every layer, from training data to runtime context to agentic commerce.
From the lab
Case studies, research, and press.
Fashion • Primary Source Datasets
Fortune 500 fashion brand builds real-time taste prediction engine.
A Fortune 500 fashion retailer used primary source datasets to infer style preference before a customer had any purchase history.
Research · October 30, 2025
Open Problems in AI Data Economics
We introduce data economics as a coherent field and define the open problems that haven’t yet been formalised. Most AI economics research focuses on downstream effects. We argue you can’t understand AI’s economic trajectory without studying how data, compute, and labour interact at the production layer.
Experiential • Brand Activation • Context Gateway
Thweet creates 1,497 one-of-one guest experiences.
Thweet turned one creative concept into 1,497 personalised experiences during Korea Blockchain Week 2025.
Press · MIT News · April 2025
Vana lets users own a piece of the AI models trained on their data
MIT News on the founding story, the Media Lab origins, and the case for user-owned AI.
AI Training Data • Context Gateway
Frontier AI Lab sources real human conversation data with consent, redaction, and provenance.
A frontier AI lab used Context Gateway to source real human conversational data with documented consent, redaction, and provenance.
Research · August 1, 2024
Model Influence Functions: Measuring Data Quality
A framework for attributing model outputs to training data contributions. The methodological foundation for pricing and valuing datasets in commercial AI pipelines.