Cases by Datapult

This is some of problems we have solved for our clients in the last year. Are they relevant for you? Reach out to info@datapult.dk. We would be glad to build similar solutions for you.

Deep Learning on Satellite Images

Deep Learning on Satellite Images

 YouTube Let's talk?
A company was interested in getting information on every household in Europe. Combining satellite images with public geocoding data and computer vision we built a serverless system that look at households all over Europe. Check the YouTube video for more details.
Neural Machine Translation

Neural Machine Translation

Let's talk?
Using the latest advances in neural machine translation we helped a publisher automate translation between four small languages. These languages are considered small by big American companies but was a massive market for this publisher. By applying the latest research in NLP we showed translation works not just for English but for smaller languages as well.
Outlier Detection

Outlier Detection

Let's talk?
This company had a team of 3 specialists to detect if their system was running smoothly. Datapult delivered a system that could automatically flag anomalies so that the specialists could look forward rather than backward in their work.
Pairing Lists

Pairing Lists

Let's talk?
A manufacturer that sold goods through retails recieved lists from each retailer that was a bit different so they had to manually pair each list from every retail. This is really boring and slow for a human but quick and fun for a computer.
Production Surveillance

Production Surveillance

Let's talk?
A big manufacturer of foods wanted to use image recognition to automate parts of their quality control. Using IoT devices Datapult built a production surveillance technology stack with AI-enabled IoT devices.
Search Engines

Search Engines

 YouTube Let's talk?
This ecommerce store wanted to sort their user searches based not on price but on what the user was likely to buy. Using a neural network and past searches we built system that ranked searches and increased conversion rates. Check the YouTube video for more details.
Stock Forecast

Stock Forecast

Let's talk?
This brick-and-mortar store wanted to predict how many units it could sell every day for the next 30 days. Using public data and previous sales data we built a model that predicted sales so the stock was not too low nor too high. We have applied the same model to predict call-center activity.
Term Planning

Term Planning

Let's talk?
Every year this school undertook the huge task of planning its terms. It was such a big task that it was only done once and as a result it was only done once a year. When we automated that task it allowed them to re-schedule the term many times a year responding to changes in demand of students and supply of teachers.