Blogs (27) >>
ICFP 2017
Sun 3 - Sat 9 September 2017 Oxford, United Kingdom
Sat 9 Sep 2017 09:35 - 10:00 at L2 - CUFP Talks 1

AppNexus is an internet technology company that powers the real-time sale and purchase of digital advertising via an auction technology called ““real-time bidding.”” A key component of our technological stack is the AppNexus Programmable Bidder which allows buyers to plug in custom algorithms in order to modify bidding behavior. Most of those algorithms are coded using the Bonsai language, a sub-Turing declarative decision tree DSL with a Python-like syntax.

Bonsai abstracts away most of the messiness of evolving industry standards. The created trees are symbolic and easy to reason about for ad-tech experts who are not software engineers. The language semantics say nothing about how trees are evaluated which gives us a lot of freedom regarding the execution strategy.

We run the models on our own hardware, in a multi-tenant environment with sub-millisecond firm real-time constraints. We also host the compiler and the model’s source code, intermediate representations, and compiled output. This makes our language design, optimization and deprecation processes very different from traditional compiler and language development. This talk will present many of the challenges that we’ve faced while developing a hosted debugger-less DSL targeted at non-programmers.

Sat 9 Sep

cufp-2017-talks
09:10 - 10:00: CUFP 2017 - CUFP Talks 1 at L2
cufp-2017-talks150494100000009:10 - 09:35
Talk
cufp-2017-talks150494250000009:35 - 10:00
Talk