Actively is building ML tools to enable every knowledge worker to be a data scientist. This is an ambitious mission; today, only a small handful of people at a small handful of companies can unlock value from the exploding amounts of data being generated. Our goal is to build interactive ML tools for every job function in an organization that combine the best of human expertise with cutting-edge machine intelligence.
Our core innovation to achieve this, built on research our co-founder conducted at Stanford in active learning, is a feedback-driven approach to model building that enables users who know the problem domain to iteratively inject their domain expertise into the system. Our "data scientist in a box" platform empowers users with no data science experience to combine their specialized domain expertise with a variety of interactive AI-backed tools (clustering, needle in a haystack, categorize datapoints, make predictions, etc.) in order to quickly answer questions about their data and create reusable AI-powered dashboards, automations, and internal tools in just a few clicks.
We’ve raised $5M with backing from world-class investors including First Round Capital (seed investors in Looker, Square, Uber, Notion, Roblox), early Stripe executives, Neo, top-tier angels like the founders of Snorkel and Ramp, the heads of ML at Kensho and Flatiron Health, and industry luminaries such as a former Federal Reserve governor and a former NBA General Manager. We’re working with multiple unicorn customers and are growing the team as we begin to scale across use cases in growth, analytics, and marketing!
What if we could build a set of tools around any knowledge worker to allow them to answer data science questions themselves? Better yet, what if these tools actually leveraged and improved from the user's expertise and feedback to specialize to their problem domain and enable even better results than a data science team could achieve, faster? By making it 10x faster for users to answer questions about their data — and to do this on their own — we'd enable them to ask more questions, understand more data, and generate more insights to drive business decisions.
This is what we are building with Actively, a no-code AI platform for business teams. Our software empowers users with no data science knowledge to combine their domain expertise with a variety of interactive, graphical AI tools in order to:
Here's how it works: we've abstracted data science workflows into a composable set of AI "primitive operations" (clustering, finding a needle in a haystack, categorizing data points, making predictions, etc.). For each of these operations, instead of requiring a data scientist to write code, we've built a rapid, iterative, feedback-driven interface around a subject matter expert to quickly specialize the abstraction to their problem, allowing them to build an accurate model to power that particular operation and surface insights within a few clicks.
See below for a demo of the core tech, as well as a deep dive on the key technology (active learning) that enables these fast, interactive AI workflows around a subject matter expert.
(This is a slightly old demo, so our current product interface looks different and we have several more “building blocks” [e.g., causal inference], but this should give you a sense of some of the tech behind the scenes!)
https://www.loom.com/share/6b481e274464478fb5182a95cd5521b0
How do we enable business users to do data science/ML work themselves and actually obtain high-quality results? Our core thesis is that active learning (an ML technique for building models significantly faster/with significantly less labeled data by selecting the few best data points to ask a human expert to label) is the key because as it makes ML & data science: