2020-03 - ongoing
Darjeeling is a highly-available system for serving machine-learning models

Responsibilities

• Architectural evaluation of available model-serving systems
• UML modeling of use cases and the microservice architecture
• Integration of Tensorflow Serving into a custom REST-API developed with Python (Flask)
• Development of a REST-API for serving scikit-learn models (Flask)
• Dockerization of individual components
• Formulation of a GitLab Runner pipeline for testing and deployment
• API documentation via OpenAPI/Swagger
• Presentation and onboarding of customers

2020-03 - ongoing
This project aims at developing a reinforcement-learning based decision-support system for train operators.

Responsibilities

• Agile software development according to Scrum
• UML modeling of components and classes
• Quality assurance during the sofware development process
• Establishment of pipelines for software and model testing
• Development of states/rewards/metrics for Ray/RLlib

2019-03 - 2020-03
Click&Ride is an application that enables railway operators to reserve routes for cargo trains.

Responsibilities

• Agile C++ software development according to SAFE/Scrum
• DevOps: Maintenance of Jenkins pipelines for testing/deployment/documentation, improving the release process
• Transitioning from database-storage to protobuf-based storage
• Implementation of new testing approaches using GUnit
• Establishment of AsciiDoc-based project documentation

2017 - 2019
geno2pheno[ngs-freq] is an application for the interpretation of viral drug resistance using next-generation sequencing data.

Responsibilities

• Development of a REST API (C++) for serving machine learning models
• Transformation of NGS data to frequency files
• UI development with Typescript and React
• Conception of new visualizations for drug resistance predictions and implementation using D3.js

2016 - 2018
An application for multiplex PCR primer design.

Responsibilities

• Development of a primer design library using R
• Development of an interactive user interface with Shiny
• Conception of a new primer design algorithm that solves a set cover problem
• Formulation of a logistic regression model for the prediction of amplification events
• Execution and reporting of primer design runs

2015 - 2016
A web service for the identification of HIV-2 coreceptor usage.

Responsibilities

• Formulation of a new support vector machine for the prediction of HIV-2 coreceptor usage
• Model evaluation using AUC-based cross-validation
• Integration of the model into a C++ backend and a PHP frontend
• Analysis of model weights for the identification of novel markers