It was built as _the_ search platform for a web-based document search and review system, providing real-time search, analytics, and high-volume updates.
Let me try to be a little less vague here (I guess to much time has passed to edit?).
In general, we developed this for searching structured (but not necessarily relational) full-text content with an enormous amount of associated metadata.
ZomboDB has come out of the legal e-Discovery world and is the backing search platform for one of the major e-Disco service providers in the industry.
It's hard to describe a typical dataset, but anywhere from 600k rows to 100M rows. Some datasets are just a few gig (on disk in Postgres) and others approach 1TB.
A typical usage pattern for a document review system is that humans (and sometimes automated computer processes) literally read each document, make judgement calls about each one, apply metadata, and move to the next document. Rinse-wash-repeat. On a large-scale litigation review, it's not uncommon to have hundreds of users doing this simultaneously.
As such, over time, every document gets UPDATEd at least once (typically 4-5 times due to administrative tasks).
You might can imagine that 100M documents with maybe 400 reviewers is a bit of an organizational problem in terms of teams of reviewers, what they should be reviewing, etc, so it's important that the data never lie. If the system says there's 1,863,462 "Potentially Privileged" documents, then that better the actual answer.
Because a system like this has to provide real-time progress reporting, analytics on judgement calls, along with the ability to generally search and organize a lot of data, we needed something that first and foremost provided transaction isolation. Enter Postgres. We also needed sophisticated full-text searching. Enter Elasticsearch.
From there it was trying to answer the question "how do we make full-text searching honor our transaction isolation rules?" That's how ZomboDB came to be.
I would think any kind of data management system that wants a canonical "source of truth" system (ie, Postgres) easily text-searchable, ZomboDB would be useful. Document Review, Medical Records Management, Inventory/Catalog Systems, etc. As I said in another post, the fact that ES is abstracted away and that need to asynchronously synchronize data into it goes away, it's fairly compelling. My hope is that some of its bigger caveats (such as crash recovery) get solved sooner rather than later.
