Hi, my name is Jörn-Philipp Lies
I'm a data & machine learning engineer.

Get in touch

About Me

Profile Image

Highly motivated data & machine learning engineer with over 10 years of experience building and deploying innovative solutions. Skilled in a wide range of machine learning techniques and programming languages. Proven ability to design, develop, and optimize complex machine learning architecture in fast-paced environments, particularly within the field of medical AI and aerospace.

If you're looking for someone who can help you define what kind of solution you need for your specific problem, implement a prototype, or develop the MVP for you and your clients, you found the right place.

Get in touch        Github profile

Projects

Retinacorder

Main product of eye2you. A medical device to examining the retina for different diseases and anatomical anomalies and to be used by non-expert medical personnel. Recording and analysis are powered by an artificial intelligence which provides feedback on the quality of the recorded image and the sufficiency of the covered retinal area. The recorded images are stitched to a panoramic retina image by taking the anatomical structure of the retina into account. The panoramic image can then be analyzed for specific diseases, anatomical anomalies, or measure anatomical features like cup-to-disc ratio of the optic nerve.

All computations are done on the smartphone using OpenCV, PyTorch, C++ and Kotlin. Models were optimized and ran in real time (<1s per model).

Kotlin badge Android badge Python badge OpenCV badge C++ badge PyTorch badge Jenkins badge Nextcloud badge
Project Image
Project Image

Bahnalyze

Continuously monitor the German train network and create a consistent and persistent database of the train delays to use for predictions.

Train time tables and current delays are only available for a few minutes in the past and a few hours in the future, which makes it difficult to use past data to predict which trains are notoriously late and which connection will likely be missed - especially annoying if you have an important meeting.

Bahnalyze collects schedule and delay data every few minutes and aggregates them. Nightly, the data of 1164 long-distance train lines is analyzed and the statistical models are updated in Firestore. This allows to provide the mean and median delay at every stop, as well as the confidence intervals to.

Future extensions in progress are predictions how likely a connection will be missed and what the expected arrival time is, given the history of the train line.

Python badge Pandas badge Docker badge Mariadb badge Jupyter badge Firebase badge Scikit-Learn badge DigitalOcean badge Scipy badge Raspberry Pi badge
Project Image
Project Image

Train prediction

Travel assistant which provides an overview over historical train delay data per train line, checks the train line the user is currently on, and provides estimates for connections.

Mobile application available for Android, iOS, and as website which provides access to the historical train delay statistics for all 1164 long-distance train lines in Germany. If you are logged into the wifi of an ICE train, it can detect the train number and show the current delay compared to the expected delay.

You can bookmark your favorite train lines for faster access and get an overview on the current status of the German train network.

Flutter badge Android badge Firebase badge Google Play badge Google AdMob badge
Project Image

Bahn Bingo

Multiplayer bingo game for incidents (delays, defects, and other annoyances) during train travel.

German trains are no longer known for punctuality and comfort, but rather countless delays, defective trains, and bad communication. Especially during peak travel times, many annoyances will happen. To make annoyances into fun, you can play Bahn Bingo with some friends.

Each player gets his/her own field, and progress is shared via Firebase. You can join the game using the game ID provided by the organizer. The game ends once a player has Bingo.

The app is currently in closed beta.

Flutter badge Android badge Firebase badge
Project Image

Fooracle

Football/Soccer cup game prediction based on Kaggle historical game data.

Using the history of all available international games back to 1872, the fooracle predicts how all games for an international tournament (world cup, europe cup, ...) will end. The historical data was used (a) to fit a probability distribution for the number of goals scored by a specific team as home or away team. For each game, a random variable is drawn from the distribution to simulate the game and (b) to train a multi-layer perceptron network to predict the outcome of any game after 90 min.

It supports the group stage with round robin mode where the 2 top teams are qualified for the round of the last n teams, depending on the number of groups, with tournament mode. While the distribution based model is non-deterministic, the neural network is deterministic and will predict the same result. To make it more realistic, you can sprinkle some fairy dust on the network, which adds some noise to the prediction.

Feel free to play around. Source code is on github below.

Python badge Pandas badge Jupyter badge Scipy badge
Project Image

Contact

Interested in working with me?

Get in touch