Saturday, March 21, 2015

Graphistry: large scale data visualization

I connected with Leo Meyerovich, for a quick overview of Graphistry.

Who is behind Graphistry?
Graphistry spun out of UC Berkeley’s Parallel Computing lab last year. It stems from my Ph.D. on the first parallel web browser (Mozilla etc. are building new browsers around those ideas) and from Matt Torok (my RA), who built Superconductor, a GPU scripting language for big interactive data visualizations. 

What does Graphistry do?
Graphistry scales and streamlines visual analysis of big graphs.  Think answering questions about people (intelligence, sales, marketing), about things (data centers, sensors), and combinations of them (e.g., financial transactions). For example, we used it to crack a 70K+ node botnet a couple days ago. Our tool immediately revealed the accounts involved, their different roles, especially key accounts, and, after 30min of interactive analysis & googling, the credit card & passport theft operation it funneled to. Most tools can only sensibly show hundreds of nodes,  and a couple open source ones handle tens of thousands, but we’re already pushing 100X more than that.

How do you use GPU?
We're taking the last 20 years of infoviz research out of papers and into accessible tools by (a) powering them with big yet economical clusters of GPUs and (b) prioritizing interaction design. The GPU side is cool. For example, our unusual backend has JavaScript orchestrating our GPU cluster via node-opencl. Likewise, we take advantage of recent breakthroughs — including our own — in optimizing irregular graph algorithms on GPUs for multiple magnitudes more data & speed.  With all this power, we're deploying atypically smart visualizations that take advantage of computationally-intensive machine learning and physics algorithms. Likewise, we're adding interactive analysis tools on top that, till now, were impossible. I can write so much here!

What is your business model?
We currently work closely with customers on big problems (contact me if this sounds relevant). We’re actively working towards self-serve analyst tools for a couple industries, and want to share our APIs with internal dev teams and analytics providers to build tools for their more unique problems.

What is your target audience?
We currently like problems in IT (e.g., making sense of activity in big networks or many endpoints) and various security problems. We're starting to expand into problems in finance (e.g., risk, fraud) and sales/marketing (social & business networks).

Can you share some demo links?
I can’t yet share the interactive versions, but here’s a screenshot:

Are you looking for funding?
As you can probably attest, startup life is intense. We’re more interested in collaborating on good problems right now. 

Are you hiring?  
Graphistry is currently 5 Berkeley engineers — a mix of language designers, compiler builders, GPU hackers, and web devs — and that’s it! We'd especially love to talk to any frontend and data viz engineers about designing big interactive visualizations & tools that were previously impossible. Consider yourself invited to our new Oakland office for an amazing show-and-tell.

Anyone interested in watching some live Graphistry demos is welcome to join our Data Science Summit, July 20 in SF.

No comments:

Post a Comment