Pareto Selection for Search Ranking
At Comtravo, amongst others, we process incoming requests with travel itineraries. One way or the another, such requests are converted to a structured format...
Building a real cross-platform app including continuous delivery
At Comtravo, we have been working on a huge Angular application for more than 4 years now. By the end of 2019 we managed to make every part of the applicatio...
Billing Engine on AWS Serverless
At Comtravo, we utilize AWS serverless infrastructure for many of our business cases. Lambda functions and Step functions, in particular, enable us to design...
Uncertainty in Recommender Systems
Bayesian neural networks are gaining popularity in the industry. In this post, we break them down and talk about how they can be useful as a recommender syst...
KONVENS'19 - The German-focused NLP Conference
KONVENS is an annual conference, which gathers together the computer scientist and computational linguist community working with the German language.
Serverless Infrastructure in the Comtravo Backend
Comtravo CTO Marko Schilde talks about scaling tech teams, investing in people, and how we’ve put serverless technologies at the heart of our business.
Our Journey from VPC Peerings to Transit Gateway
At Comtravo, we run our services in the cloud with AWS. We run multiple environments each having their own VPCs. We have been using VPC peering connections t...
Machine Learning on `dask`
The data science team at Comtravo uses dask to coordinate fairly complex machine learning workloads, both for training and running them in production. Our NL...
Effective Mocking of Unit Tests for Machine Learning
The data science backend at Comtravo is almost exclusively written in Python (3). We currently have about 10 different machine learning models performing var...
Async/Await and JavaScript Arrays
The Comtravo backend is built from a number of micro-services, most of which run on Node.js. Like any large scale business application we need to do a lot of...
PyData Berlin Geocoding / Mapping Special
We kicked off November with a PyData Berlin meetup. Pizza, drinks and festive Spekulatius, along with three great speakers of course, guaranteed a great turn...
Named Entity Recognition using Neural Networks
Named-Entity Recognition (NER) is an NLP task that involves finding and classifying sequences of words (tokens) into pre-defined categories. Examples of name...
A Practical Parser for Time Expressions
One of the main methods our customers communicate with us is email. We get hundreds of travel booking emails daily; all of them contain time expressions that...