Accurate. Honest. Rich. Introducing Filtered's search API.

By Toby Harris

2 minute read

This week we shared the credentials to access Filtered’s new search API with our first customer. This customer will now be able to send any kind of query to us and receive back polished metadata for the most relevant items in their catalogue, including relevance scores. Why is Filtered’s search needed for this customer?

Here’s the use case:



  • The LXP is a multi-tenant platform with 100s of customers
  • Digital content available has a cost associated with it (and a huge ROI vs in-person training!)
  • Giving access to all content to all users is not helpful or cost effective
  • Instead, courses and resources are actively curated in to target certain skills
  • The customer has been using Filtered to browse the universe of content available
  • Our search API means this can be put into a simple front end for customer admins
  • ie, they can now self-curate new courses and resources

What value does Filtered’s search API add here?

Well, it’s simply the most accurate search in the industry by any benchmark. Our natural language algorithm is one half of a language model (ie, it just encodes text as numbers, and doesn’t ‘decode’ them back into generative outputs) which has been optimised to produce results as good as a human expert. The results are also rich. We put a lot of work into making sure your metadata is in tip top shape (for example, we simply won’t accept content with unusable descriptions) and we make it easy to know how long something will take to consume and the specific format it comes in. Our search is also honest. Uniquely, we also tell you how confident we are in the results via relevance scores:

For example, the query ‘machine’...

Screenshot 2024-11-21 at 13.44.01

… might return this result, where the relevance score of 0.599 suggests that it’s good, but not a perfect match:

Screenshot 2024-11-21 at 13.46.10

As a user, you can apply your own judgement even without being an expert. You know when there’s something good here and when there isn’t. It’s much like the way AI seamlessly augments everything from our experience of travel (looking at estimated travel times to see if a day out is feasible) to hospitality booking (looking at price + rating to see if a meal out in that city is worth it).

Now this benefit of data-informed learning choices is available to applications, to craft the user experience...

Rather than blindly surface everything tagged against a query, we can choose to surface things depending on the frame of reference. If we’re trying to tag content we want to see lots of results. But, if we’re looking for career-related learning recommendations, we need hyper-relevance.

The Filtered browsing interface uses the same functionality to be clear about whether there really is anything hyper relevant to a skill or not:

In this case there's highly relevant content available for 'GenAI' and 'Managing AI projects' but less so for 'Giving an AI product demo'.

The problem of fire-hosing users with a deluge of irrelevant results has always been there in learning platforms. And now it’s back with a vengeance as the new breed of talent intelligence systems are also beset with the same issue.

It can seem like a long time between building something, testing it in multiple sandboxes (you can see the Filtered API at work in our AI agent here) and delivering it as a contracted item.

But now we have customers for this new service under our belt, we are ready to take Filtered’s laser-guided, transparently accurate search to end users wherever they are in need.

Book a demo

Free learning content library benchmark
Filtered logo rotating

Get the best return on your L&D spend.