At the end of 2019, my co-founders and I started Baseten to solve many of the issues that we’d faced building machine learning models at companies big and small, as ICs and as leads.
Before we started Baseten, we spent a lot of time talking to data science and ML teams about the frustratingly slow pace of homegrown machine learning efforts and the barriers that stopped most of them getting to value.
Everyone was already onboard with machine learning at a high level; the data engineering landscape started to mature very quickly making machine learning more tenable, and the breakneck speed of research driven by the ever decreasing cost of compute meant that training effective models wasn’t the bottleneck.
Instead it was the small things. Deploying models. Wrangling Docker. What’s Kubernetes? Integrating data in. Pushing data out. Fighting for resources to build some UI around the model. What we saw was that while many teams had set out to get value from ML, most were ill-equipped from a skills, resources and credibility perspective.
We aspired to build the general purpose toolkit that we wished we had — a product to leverage a technically-savvy, scrappy data scientist’s skills and turn them into full-stack shippers. We wanted to bundle and productize the things we struggled doing the most: deploying models, setting up backend infrastructure to integrate those models, and building UIs to expose predictions in a given business context. We envisioned modular pillars so you could integrate in parts, but using them all together would unlock feedback loops that we hadn’t seen many machine learning teams achieve.
Shipping our public beta
After over 18 months of building, we quietly announced Baseten last May. The product was raw (and clunky), but we wanted to prove our hypothesis that a bundled approach to shipping machine learning projects could in fact work. The results of this experiment have been fantastic — we’ve worked with users and organizations of all shapes and sizes, and are more convinced than ever that we can dramatically lower the cost of using machine learning by providing a set of building blocks for each step of the journey.
With the help of an inspiring set of early users, and after some dramatic product iterations where we painfully rethought and rewrote core abstractions, we’re excited to launch the public beta of Baseten to allow data science and machine learning teams to build full-stack applications without having to become infrastructure, backend, and frontend engineers.
With Baseten, you can:
- Serve — Baseten’s model deployment mechanism allows you to get models served in minutes and provides an intuitive UI to manage, monitor, and configure infrastructure when needed.
- Integrate — Embed your models in business logic and serve them behind serverless, scalable APIs without configuring any infrastructure.
- Ship — By drastically reducing the barriers to getting models out of notebooks and into the hands of users, the rate of iteration drastically increases, which in turn helps the business derive more value from machine learning.
We raised some money!
We’ve raised a little over $20 million dollars to date across our seed and Series A rounds.
Our Series A was led by Sarah Guo from Greylock Partners who also co-led our seed round in 2019. We’ve been lucky enough to know Sarah since we started at our first company almost 8 years ago. She’s always been a source of energetic, thoughtful discussions and when we set out to start Baseten, we knew we wanted her to be a part of this journey.
We’re also thrilled to have the team at South Park Commons, Lachy Groom, and Ray Tonsing (Caffeinated Capital) invest alongside a group of outstanding angels including Greg Brockman, Dylan Field, Pascal Levy-Garboua, Mustafa Suleyman, DJ Patil, Dev Ittycheria, Jay Simons, Jean-Denis Greze, Peter Skomoroch, Sandy Kory, Alana Anderson, and Cristina Cordova.
Our vision is to amplify and accelerate the impact that machine learning has on how we live and work. To realize this tall order, we’re actively hiring for roles across engineering, design, and go-to-market.
We’d also love to have you try the product; we can’t wait to see what you build!
Fine-tune FLAN-T5 on Blueprint today!
You can now fine-tune FLAN-T5, an instruction-tuned text-to-text transformer model developed by Google, on Blueprint!