Urbint

We have a very flat engineering organization, with the two primary focuses being data infrastructure and machine learning. We practice agile development with quick stand-ups every morning and sprint planning every two weeks.

We have a logical unification engine which we use to determine whether or not two things are the same. For example, if we see a person named “John Smith” on two different documents that we ingest, how do we determine that they’re talking about the same person? What if we see “Jon Smith” - is that the same person, or a typo? While many in our industry use string equality, that method is prone to both losing information and being overly presumptive. Our unification engine gives us a framework to explore these puzzles.

We process a lot of data and typically work on the BEAM (the Erlang / Elixir VM) which is highly concurrent and distributed. Determining how to archive our task concurrently can be the difference between minutes and days when processing our datasets.

Lastly, to achieve the performance characteristics needed to process our data and apply our algorithms, we wrote our own (graph) database. Having this level of vertical integration enables us to tune every aspect of our stack and lets us do things that would not otherwise be feasible.