Rosenheim, Bavaria, Germany
-Time Series Forecasting experience with neural networks and TensorFlow 2.0. Rank 1 solution at AICup University of Passau Trading Hub Europe Gas Prediction Task. -Machine Learning experience in Python, PyTorch, HuggingFace, etc. through in-house NLP research projects since September 2021. Prompt-Engineering, experiments and communication with GPT-3 with in-house backend services. -Front-end (native iOS & Android, Flutter), back-end (Java and Spring Framework) and DevOps experience (Docker, git integrations) through activities in the company since October 2014.
- Performed data preprocessing and visualization - Evaluated various statistical and ml-solutions for time series prediction in the domestic heating domain - Designed reproducible and clean code project structure for data science workflows and components - Implemented the AI use cases with a deep learning library & implemented components - Containerized and deployed ml components - Delivered various presentations and pitches during competitions such as AI-Cup and Rosenheimer Gründerpreis - Maintained contact and conducted workshops with customers related to AI use cases
- Built automatic end-to-end pipeline for data evaluation, preprocessing, training (continuous training) and metadata aggregation - Evaluated tools like KubeFlow, DVC, MLFlow etc. for this particular task
- Gained knowledge about the project by interviewing developers and analyzing project - Identified opportunities for optimization in data science processes
- Evaluated different AI solutions to aid interpretation of interview texts such as Sentence BERT for comparing/clustering similar text passages, BERT for sentiment analysis, and T5 for text-to-text automatic interpretation - Trained open-source Transformer-based LM (T5) for automatic interpretation of interview texts & Fine-tuned GPT-3 Davinci Model for this particular task & compared GPT-3 results with on-premis solutions by using adequate metrics - Performed data/text-preprocessing tasks - Used platforms like Haystack to build a knowledge base search system for scientific interviewers - Designed and implemented AI text-to-text webservice that automatically injects prompts to user input based on the user intention and used GPT-3 and Aleph Alpha LMs to fullfil the task
- Constructed CI/CD pipelines for multiple projects - Containerized services and apps using Docker for streamlined deployment - Proposed and implemented software architecture patterns (e.g., Repository, BLoC), created project templates for future use, and conducted presentations to educate colleagues on pattern usage - Refactored services to microservice architecture and integrated ELK stack and GrayLog logging - Established KeyCloak‑based authentication infrastructure and integrated it into apps - Managed infrastructure setup and maintenance
- Started with implementing new features into existing Android apps using Java - Set up database schemas for new projects and features in MySQL - Developed and adjusted existing webservices - Started a new iOS app with Objective-C, then continued development in Swift - Developed a mobile 2D platformer game using the Unity game engine and C#