Dust improves team productivity with large language models on internal data

Dust, a French AI startup, breaks down internal silos, surfaces important knowledge, and provides tools to build custom internal apps to boost team productivity. Dust uses LLMs on internal company data to give team members superpowers.

Gabriel Hubert and Stanislas Polu co-founded the company. In 2015, Stripe bought their startup, Totems. They left Stripe after a few years.

While Gabriel Hubert led product at Alan, Stanislas Polu worked on LLM reasoning at OpenAI for three years.

Dust was another collaboration. Dust doesn’t build large language models like other AI startups. Instead, it wants to build apps on OpenAI, Cohere, AI21, etc. LLMs.

A platform for designing and deploying large language-model apps was developed first. It then focused on one use case—centralizing and indexing internal data for LLMs.

From internal ChatGPT to next-gen software
Notion, Slack, Github, and Google Drive have several connectors that pull in internal data. Indexing this data allows semantic search queries. A Dust-powered app finds the relevant internal data, uses it as the context of an LLM, and returns an answer.

Say you joined a company and are working on a long-running project. Find information in internal data if your company promotes communication transparency. The internal knowledge base may be outdated. An archived Slack channel may make it hard to understand why something is done this way.

Dust goes beyond internal search results. It can search multiple data sources and format answers in a more useful way. It can be used as a ChatGPT or to create new internal tools.

“We’re convinced that natural language interfaces will disrupt software,” Gabriel Hubert told me. “In five years, it would be disappointing if you still had to click on edit, settings, or preferences to change your software’s behavior. Because you, your team, and your company are unique, we see more of our software adapting to your needs.

Design partners are helping the company implement and package Dust. Stanislas Polu told me that enterprise data, knowledge workers, and models can be used to support them.

Dust is tackling an intriguing issue in its early stages. Data retention, hallucinations, and LLM issues pose many challenges. LLMs may reduce hallucinations. Dust may create an LLM for data privacy.

Dust has raised $5.5 million (€5 million) in a seed round led by Sequoia, with XYZ, GG1, Seedcamp, Connect, Motier Ventures, Tiny Supercomputer, AI Grant, and a number of business angels, including Olivier Pomel from Datadog, Julien Codorniou and Julien Chaumond from Hugging Face, Mathilde Collin from Front, Charles Gorintin and Jean-Charles Samuelian-Werve from Alan, Eléonore Crespo and Romain Niccoli from Pigment, Nicolas Brusson from BlaBlaCar, Howie

Dust believes LLMs will revolutionize business. Dust works better in a company that values radical transparency over information retention, written communication over endless meetings, and autonomy over top-down management.

LLMs will boost productivity if they unlock knowledge workers’ potential, giving some companies an unfair advantage.