In May 2017 our first Implisense Hackathon took place. Where? On the sunny island of Mallorca. We booked a finca owned by nice people from London and spent a week working on creative coding. Beside  cooking together, hands-on-traninig in Go, fitness training, we worked on four challenges:

  • Automated predictions
  • Proximity Geo-Analysis
  • Chatbots for complex searches
  • Image annotation for Facial Keypoints

In the middle of the night

Automated predictions

Time series data plays a major role in the Big Data phenomenon. At the same time, the tools have become better in dealing with massive time series. We were curious to know if new open source tools could be used to predict future management changes at scale. This idea came from discussions with some customers who were able to place their infrastructure projects whenever a change of key management personnel took place six months ago. How well can time series data from management changes be used for sales related predictions? We found out it was possible to predict the month, week and volume of expected management changes for selected industries with justifiable estimation errors. The team discussed then how to combine these predictions with recent developments for a given company to deliver account specific pre alerts.
The team also tested whether the approach is suitable for predicting newly available content from websites. This would allow us to optimize the utilization of our IT and achieve a higher degree of up-to-dateness on the customer side. Team Operations integrates these developments into the production system for 2018. Open jobs here.

Proximity Geo-Analysis

Many product buyings have a sublime spatial reference. But for many vendors complex geo-data analysis would not be feasible. Freight forwarders e.g. usually settle near motorway exits, ports or airports. Management consultants are usually located in the city centre or near airports or high-speed train stations. If geographic proximity to specific points of interest (POIs) plays a role in the implementation of business models, how can we integrate these characteristics into our customer recommendations? A team has developed a component to calculate the routing distances per company location to selected points of interest from open street maps, such as the proximity to the nearest motorway access road or the distance to the nearest industrial park. Exemplary analyses have shown which unexpected challenges arise during the integration of OSM data and which geo-data should rather be obtained by commercial providers in order to achieve a higher reliability in the significance. The newly developed geodata is included as additional data points in our Company Index. The responsible colleagues in the Data Analytics team are looking for additional employees for further geodata analyses.

Chatbots für complex searches

Chatbots promise – after the fulminant awake of Siri, Alexa and Co – natural interactions beyond the classical interfaces in the business area also for professional users. We asked ourselves which complex searches could be realized more efficiently via chatbots. Could we develop our own chat bots based on our API? Could they be integrated into popular Messenger systems such as Whatsapp, Telegram or Slack? This challenge was taken up by two teams and until the end of the week a first chatbot was developed to allow company selections using faceted full-text searches. These developments are being incorporated into a new service that will be tested in 2018 with pilot customers from the Enterprise segment.

Image Annotation for Facial Keypoints

An interesting side issue was dealt with by a team in the field of visual analytics. Our open source tool Dalphi was developed to generate training data for machine learning processes. So far, we have used it primarily for analyzing mass text data. From the feedback from other startups we have learned that the generation of training examples is also time-consuming for the automated analysis of mass image data. The team has decided to develop an appropriate annotation interface for the hackathon and to use it for the recognition of features in faces (facial keypoints). As a backend service, a Convolutional Neural Network (CNN) consisting of annotated images should be trained in order to obtain a model for keypointer recognition. At the end of the hackathon, models were trained to recognize certain people in photos. We will discuss these new analytical skills with our key accounts to find the best practical use case for them. Useful. We were able to sort all photos of the Mallorca Hackathon according to the recognizable persons without having to ask Google or Apple.

The next day we went to Playa de Formentor to enjoy the natural beauty of Mallorca


We worked on four challenging ideas in five days. We examined their technical feasibility. Mallorca is an ideal place for that, especially in early May. The island is very easy to reach from Germany. There are inexpensive properties with low season rates and the places and beaches are not very crowded.

We organize more hackathons:  The next one is already planned in the desert of North Africa. Join the team? Job offers can be found here: