Consultant for AI and Machine Learning
Results-driven AI and Machine Learning Engineer Consultant with a proven track record in leading and executing cutting-edge AI solutions that drive business transformation.
Expert in developing and deploying sophisticated machine learning models using technologies like Python, PyTorch, TensorFlow, Keras, and scikit-learn, with a deep focus on generative AI, Computer Vision, and anomaly detection.
Demonstrated leadership in managing cross-functional teams, driving AI project lifecycles, and implementing CI/CD pipelines, ensuring the seamless integration and scalability of AI solutions.
Extensive experience with cloud platforms (AWS, Azure), containerization (Docker, Kubernetes), and distributed computing, leveraging large-scale data to optimize business processes and decision-making.
Proven ability to take ownership of AI initiatives, from conceptualization to deployment, and mentor teams to achieve technical excellence.
Adept at aligning AI strategy with organizational goals to deliver impactful, business-critical results.
Machine Learning Engineer at KUNGFU.AI
Developed and optimized a Retrieval-Augmented Generation (RAG) LLM application to automate contract analysis for a Fortune 500 infrastructure services client, leveraging advanced NLP techniques and vector-based search to improve clause detection accuracy and reduce manual legal review time.
Artificial Intelligence Engineer at AerSale, Inc.
Designed and implemented a CI/CD-enabled system integrating Retrieval-Augmented Generation (RAG), fine-tuned LLMs, Pinecone vector database, and LangChain to extract information from engine documents in a generative AI-powered chatbot. This solution significantly improved efficiency for internal users involved in engine buying processes by enabling seamless information retrieval and contextual understanding. Developed a machine learning model leveraging internal engine component data and external sources to predict component demand, optimizing inventory strategies and enhancing decision-making. Additionally, utilized Power BI and SQL to create a comprehensive report integrating BTS flight and weather data, demonstrating potential cost savings for airlines through AerSaleās AerAware product, ultimately driving increased revenue by showcasing its impact on pilot vision and operational efficiency.
Machine Learning Engineer at Robert Bosch GmbH
Developed a Deep Learning-based process monitoring system using an LSTM-Autoencoder for real-time anomaly detection in multivariate sensor data, enhancing equipment availability and preventing damage to expensive workpieces. The model was extended to monitor multiple identical CNC machines, achieving an average F1-score of 0.96 across various scenarios. Deployment was streamlined using AWS SageMaker for model training, AWS IoT for data ingestion, and CI/CD pipelines with AWS CodePipeline to ensure continuous improvement and deployment efficiency.
Deep Learning Engineer at Bosch Center for Artificial Intelligence (BCAI)
Developed cutting-edge loss weighting methods for multi-task learning in autonomous driving, achieving superior performance over industry benchmarks. Extensive evaluations were conducted on scene understanding tasks, including semantic segmentation, object detection, depth estimation, and surface normal estimation, leveraging datasets like Cityscapes and NYUv2 to establish a robust evaluation framework. These novel methods were seamlessly integrated into the existing autonomous driving software, ensuring compatibility and performance optimization. The research was published in GCPR 2024, contributing to advancements in neural networks and autonomous driving, with two registered patents recognizing the innovation and solidifying its intellectual property value.
Data Engineer at Scalefree International GmbH
Led the internal Business Intelligence development team as a Scrum Master, fostering efficient and collaborative processes for successful BI initiatives. Designed and implemented streamlined data ingestion workflows for the staging area and raw data vault using AWS services such as S3, Lambda, and Batch, improving efficiency and reliability. Collaborated on a Tableau-based data visualization project to analyze customer engagement and optimize marketing strategies for a leading e-commerce client, contributing to a 15% increase in conversion rates within six months.
Student Research Project at University of Hildesheim
Conducted image-to-image translation between the domains of regular images and artworks with Deep Generative Adversarial Networks. Enhanced CycleGAN by introducing a two-objective discriminator as regularization, incorporating an adversarial self-defense for better cycle-consistency, and applying differentiable augmentation on the target domain with less data. Employed agile intercultural project management techniques to manage the project successfully.
Web App for Generating Art Pieces
Designed and developed a Flask-based web application to deploy a trained CycleGAN model on a Monet dataset. Enabled users to generate art pieces based on their input images.
End-to-End ML Framework with Continuous Delivery Pipeline
Implemented a comprehensive framework for developing, training, and deploying machine learning models. This end-to-end ML pipeline includes data ingestion, transformation, and model training as employed in industry settings. Deployable with a Flask-based web application for predictions in conjunction with AWS Beanstalk and AWS Codepipeline to obtain a continuous delivery pipeline.
Reinforcement Learning with Q-Learning for Mountain Car Environment
Implemented the Q-learning algorithm in Python with NumPy and applied it to the Mountain Car environment using the OpenAI Gym library.
Analytical uncertainty-based loss weighting in multi-task learning
In Proceedings of GCPR 2024
Proposed a novel uncertainty-based task-weighting method that optimally balances task losses in multi-task learning, achieving high performance with lower computational cost compared to brute-force approaches.
Read on arXivComputer-Implemented Method for Multi-Task Machine Learning
Patent No. DE 102023201578 (Issued 2024)
Developed a method for dynamically activating and deactivating neurons in a multi-task learning neural network, enabling task-specific loss weighting based on uncertainty measures such as variance and entropy to improve training efficiency and performance.
View PatentUniversity of Hildesheim
2020-2023
University of Applied Sciences and Arts Hannover
2016-2020
simon.kutsche@gmail.com