Last month Acrotrend beat off around 50 competitors from 10 other organisations to win the Looker London Hackathon, which was aimed at building innovative solutions using Looker.

What is Looker?

Looker is a data & analytics platform for enterprises. It helps companies analyse and visualise data to enable data-driven decisions. Click here to understand more about Looker.

Our solution.

The Looker team organised the London hackathon for Looker customers and partners with the objective of:

  • building either something new,
  • extending a functionality, or
  • telling an interesting story from data.

From the moment we heard about the hackathon we were excited to compete in it. Our goal was to build something new and add a functionality that will help end-users understand/ interpret/ communicate data driven insights.

After generating a few initial ideas we decided to build a capability for asking questions on Looker. This was primarily because, based on our experience, we have seen that business executives like the KPI dashboards and reports, but then get stuck in the “So What?”, “What do I do next?” or “How do I action this insight?” phase.

The solution we developed addressed this by giving business users:

  • an ability to directly communicate with data,
  • ask the questions in their own words without the need to code, and
  • make better decisions in that very instant.
Got 2 minutes? Watch this demo to see how our solution worked.

This is a simple Looker dashboard embedded in a web page. On the top we created a panel to ask questions in a natural language.

Let’s say I have a question after looking at this dashboard where I want to understand: What is the distribution of age group by trip count?

All I need to do is type my question in the top panel and I get a visualisation with an answer.

If we try asking a different question in a conversational way: Give me the trip duration by station ID, this time I get a table of results.

How we worked our magic.

We formed the basic foundation using LookML files attached to a project.

These files can be views, explores, etc.

This serves as a meta for our querying engine.

The provided search query by the user can be parsed in multiple way depending upon the scope of ML model. Since Looker has already provided us with an intelligent way of querying data using REST API we scoped our development to prepare payload for that querying API.

A generic payload might look like:

{‘model’: ‘bike_trips_ask_data’, ‘view’: ‘trip’, ‘fields’: [‘trip.trip_count’, ‘trip.age_group’], ‘filters’: {}, ‘sorts’: [‘trip.trip_count’], ‘limit’: ’10’}

We are using query keys as targets to classifying predicted entities in the user search query. When the user enters their question, the query predicts model, view, fields, filters and sort entities and maps it to the Looker query model. While predicting entities, LookML meta that we created in the beginning becomes useful.

The Looker API is the backbone of our tool. We leveraged Looker’s querying APIs on top of our ML model to bring magic back to the users.

The real challenge was that we had about 6-8 hours to develop the prototype and we wanted to ensure that we cover all basics and not waste time going too deep into a particular feature.

Ask more from your data.

It is interesting how this can be helpful in scenarios where you want to ask questions directly to your data even if the data points are not present in the dashboard.

If you’d like to find out more about our Natural Language interface for Looker, or participate in the beta programme, please register your interest using the form below: