TechScreen Revised

Problem: for my bachelor thesis I developed a simple Java RESTful Web Service that could be used to control a mBot robot car using the internet. For this I used a Raspberry Pi to run the web service. The Raspberry Pi was connected via USB to the mBot. The embedded mBot controls (e.g. to move forwards/backwards, spin, retreive sensor informations) were called via the internet using HTTP endpoints.

If you are not using Facebook on a regularar basis and / or you don't want to install the facebook apps on your phone, you can't access the facebook messenger via the mobile web version of the site. If you click the messenger, the site tells you to download the facebook messenger app. As I knew that it is working without an app on a desktop, I decided to try the option in the browser of the phone to show the desktop version of the webiste. Even though the navigation is not that comfortably as with the web version, you are able to access the facebook messenger.

Captcha (completely automated public Turing test to tell computers and humans apart) is a way to differntiate humans from computers. They are widley used on the internet to prevent a website from spam and bots. At a certain time during an ordering process for a sneaker, you needed to solve a captcha to get access to the plattform. Everyone needs to solve this capture and this takes up to 10 seconds. Exaclty this time can decide whether you are able to order the sneaker or not. Therefore I needed a way to bypass this captcha and gain an advantage over all other participants.

There was a public WIFI with no password protection. Everone could connect to the WIFI, but there was no internet access, even tough they promoted it at the entrance of the restaurant. No employee was able to support, so I decided to connect to the router and try the standard login credentials. Usually they are user:admin pw:admin or blank password. This worked and I was able to restart the router. After the restart, the router could connect to the internet and everyone could access the web again.

Problem: Collaborate working with IntelliJ and Java: my colleagues and I used Java and the IntelliJ IDE for a university project (online voting system). To work together we decided to use git and had the issue, that by default the git Repository was filled with too much build data, because the .idea folder of the IntelliJ project, which contains project specific settings files and per-project details such as VCS mapping and run and debug configurations, as well as per-user details, such as currently open files, navigation history and currently selected configuration.

Problem: The classification time of a trained neural network was too slow on average for a real time service. Solution: 1. Build additional, less complex models with faster inference times (eg. SVM, Decision Trees) to handle the most common requests very fast and efficient. 2. Distill the neural network (https://arxiv.org/pdf/1711.09784.pdf, https://arxiv.org/abs/1910.01108) into a smaller faster model. 3.

Problem: The request response time for a REST service we provided was too slow. Solution: Return data to API customer first and persist data later. Solution explanation: The runtime analysis showed that the write operation to the database was the slowest part of the API call. The solution for speedup was to write the data to the database AFTER returning the response to the client. While this increases the risk of having "old" data in the database for another call, that needs this data (very improbable case) it improves the response performance significantly.

Problem: Find a software framework for a small social app startup with little in house technical expertise to create a relatively simple app for Android and IOS fast, with little cost but scalability. Solution: Use Vue-Native, Google Firestore and Google Cloud Functions. Solution Explanation: The biggest/cheapest group of developers is assumed to be web developers. They are assumed to know HTML, CSS and Javascript.

A single-page web application was being served by a web server and was supposed to interact with another web server to load content like posts, comments and ratings. Although the web application could be loaded and the overall structure was visible, no content was being displayed. Looking at the console with Firefox's inspector tool revealed that the request to the other web server, an API proxy, was blocked by Firefox because the request was cross-origin and the other web server did not explicitly allow such a request.

A search function for a management software was supposed to be able to lookup a customers data by searching with their address. Addresses can change and if for example a customer moved and thus their address changed, it should still be possible to find said customer with their old address. This functionality was required for some other entities as well, so implementing it generally was favorable. Instead of explicitly modeling for every such entity that it can have multiple values over time, I used an extensions for the Hibernate persistence framework called Envers for auditing tables.