Building a data pipeline, connecting to machine learning models, setting up API and search databases
What We're Building
Customers and relations are the most valuable assets of any company.
But at closer look the data used to identify these customers is almost
always a mess. It is incomplete, out of date, not as well structured as
we need, spread out over multiple “silos”, and containing a
significantly amount of smaller and larger errors. These errors can
almost always be fixed. It is the sheer number of records that need
checking and fixing that makes for a task of Herculean proportions.
It is Frosha’s mission to solve this problem by introducing Deep
Learning to this field. We train Deep Learning models to clean and
structure customer data, to identify duplicates and relations, and merge
and migrate databases. Our system can identify “J. Smith”, “Smith,
Joan” and “Ms Smiith [sic]” as referring to the same person, even though
the records reside in different databases and have completely different
structure. Such a task quickly gets out of hand using traditional rule
based or manual approaches.