• Deployed and optimized DeepSeek LLM on a private, on-premise server, ensuring 100% data privacy for the client.
• Integrated with Open-WebUI for an intuitive user interface and streamlined model interaction.
• Increased inference speed from 3.5 to 10.2 tokens per second through GPU acceleration and model quantization.

• Developed a fully conversational chatbot, reducing customer reliance on hotline support by integrating an LLM with a custom RAG solution that combines semantic and keyword search for optimal response accuracy.
• Implemented smart email routing for ~10k emails per day, significantly reducing customer wait times and improving response efficiency.
• Optimized phone hotline routing, increasing the correct call assignment rate from 75% to 94%, enhancing customer service operations.
• Built multiple AI services, including:
  • Automated response proposal system for customer emails, improving agent efficiency.
  • Information extraction model for scanned documents, streamlining data processing.
  • PII masking system, ensuring compliance with data privacy regulations.
  • Phishing detection service, identifying suspicious emails to enhance cybersecurity.
  • Complaint classification model, enabling faster and more accurate issue resolution.
• Supervised and mentored junior employees, onboarded and educated 10+ junior colleagues across multiple projects, fostering team growth and expertise.
• Led end-to-end machine learning projects, overseeing the full lifecycle from data preprocessing and model development to deployment and monitoring, ensuring business impact and scalability.
• Conducted lectures and training sessions on machine learning and deep learning, sharing expertise and advancing industry knowledge.

• Build and integrated AI-modules to speed up the work of the editors.
• Person detection and identification of 1600 celebrities in total.
• Gender classification 95+% accuracy.
• Scene/concept classification (e.g., outdoor shot, indoor shot, underwater, night shot, aerial shot, time lapse shot)
• Sentiment analysis in text.
• Determination of spoken language in Video/Audio data.

• Trained and integrated models for virtual jewelry try-on application.
• Hand and person (face, hair, clothes) detection and segmentation.
• Algorithm performance improvement.
• Data augmentation, parameter tuning, (re)training of the algorithms.

• Custumized VGG16 neural network into a FCN to identify a road on the image.
• Trained on Kitti Road dataset

• Build and trained a deep neural network using TensorFlow,
  able to distinguish 43 different traffic sign types.
• Final model had 90% accuracy after few hours.
• Dataset: URL on request
• section=gtsrb&subsection=dataset

• Build and trained a deep neural network using Keras, which is able to predict steering angle.
• Final model successful keep the car on track whole time in the simulator.

• Setup multiple learning-management-systems (Canvas LMS, Open edX).
• Jupyter notebook integration.
• Setup infrastructure with Docker.

URL on request

• Writing software tests

• Front- & backend development for an Intranet application.
• Design and development of different software modules.