People often ask me if I’m a front-end developer or a back-end developer. I’m always glad to answer I’m both, or neither depending on your point of view. My issue with specializing in either one of them is that you lose perspective on the bigger picture. You’re no longer need to think about what happens on the other side of the fence.

When I set out to build an app where you can review music and recommend music to your friends, I thought I’d try a different approach. My quest became to share as much code between front-end and back-end as good software design would allow. I heard about triple stores and how they can store data in a very loosely defined structure without losing the ability to do complicated queries.

Hats off to the people in the Clojure community who’ve been advocating the use of triple stores (Rich Hickey, David Nolen, Nikita Prokopov). Without them I would have never even thought of using triple stores. Furthermore, Clojure makes it ridiculously simple to share code between server and client, which was essential to make this whole idea work.

Triple stores

A triple store consists of a set of triples. A triple always consists of:

  1. A subject (a user, a piece of music, a review etc.)
  2. A relation (something that the subject has or refers to)
  3. A value (this can be another subject, a string, or any other value for that matter)

This is a triple:

frank likes pizza

Where frank is the subject, likes is the relation, and pizza is the value.

Consider this database:

frank name "Frank"
frank password "passw0rd"
frank email ""

review-1 author frank
review-1 title "Glasper's killing it!"
review-1 record double-booked
review-2 author frank
review-2 title "Ryuichi Sakamoto - async"
review-2 record async

double-booked artist "Robert Glasper"
double-booked album "Double Booked"

async artist "Ryuichi Sakamoto"
async album "async"

We can now query for example all authors that wrote a review, like so:

?subject author ?author

Every word starting with ? is something we don’t know and want the database to look up for us. Words not starting with a ? are things we do know. Things we do know can be literal strings like author but also variables from our program code.

The query gets us all reviews and their authors, in this case review-1 author frank, review-2 author frank.

Taking it a step further we can also get the authors that wrote a review for Robert Glasper’s album Double Booked:

?review record double-booked
?review author ?author
?author name ?author-name
?author email ?email

As you can see the set of triples form a graph where subjects are connected to one another. If you know SQL you can see we can do implicit joins with the query language very easily. We get all reviews for double-booked and then extract the author’s name and email for each of the reviews.

Having such a loosely defined structure while retaining a very rich query capability got me thinking: what if we would use only one data structure for our entire application? The entire application state as a triple store, front-end and back-end. This would solve two fundamental problems:

  1. The data formatting problem: where the data is never formatted in the right way. How often are we not transforming our data structure to fit our environment? We would like to get rid of this glue code once and for all.
  2. The code sharing problem: Code between back-end and front-end almost never gets shared because the data structures are quite different. For rendering we often need different structure than for our domain logic.

The back-end

The back-end is set up as a machine that processes incoming user commands, validates them, and updates the application database.

An incoming command is simply a set of triples with a single label attached to it. For example a label review and a set of triples (review-1 title "Robert Glasper - Double Booked" etc..).

Since the database is a triple store we can easily share it with the front-end. First of all, the back-end sends all its triples to the front-end initially (filtering out the user passwords and other sensitive data first). Each user command can result in zero or more domain events which follow the same structure of label+triples. It also sends off these events to the front-end so that it can update its state accordingly.

The front-end

The front-end also has a triple store and receives updates from the back-end through events. However, this isn’t enough: the front-end needs more information to do rendering. For example it needs to know about UI state: which elements are highlighted etc. The front-end extends the back-end triple store with some more triples, like:

review is-highlighted true


review is-expanded true

Due to the simple structure of the triple store (after all it’s just a flat unordered set of triples) there’s no need to worry something might not fit. We can just add more triples (or remove them). The domain between the back-end and front-end is shared, so it’s very unlikely the triples in the back-end database don’t match the ones needed by the front-end. Some triples that are used in the back-end might never get used in the front-end. It’s no problem to just ignore them, they won’t break anything.

The most beautiful thing is that no matter how complicated the UI gets, we can always write a query that gives us the right data back. We can get all music reviews and their authors, and the bands that played the music etc. No need to dig into these deeply nested data structures anymore to find what we’re looking for.


In essence triple stores give us the ability to write many functions for a single data structure. This gives us a tremendous reduction in complexity and a great increase in code re-use.

While there’s a lot of novelty in this approach and the wrinkles are by no means ironed out, I can only dream of a future where application development is done this way.

To all developers I would like to say: always be critical of your application’s architecture. Never just take it for granted. Keep looking for alternatives. Write down what annoys you. But most of all remember that the front-end and back-end divide is just made up. It doesn’t need to be there if you don’t want to.