AI & Machine Learning Consultant

**Senior Machine Learning Engineer** with 4+ years of experience developing and deploying production-grade AI solutions for industrial automation, healthcare, and enterprise applications.

**Current Position: Machine Learning Engineer at Keyspot AI (October 2024 - Present)**
• Designed and implemented state-of-the-art machine learning models for automated detection of technical modifications in RFQ/RFP documents, surpassing existing benchmarks in both speed and accuracy, significantly reducing manual review requirements
• Developed and deployed multimodal deep learning architectures capable of extracting structured data from complex, nested technical datasheets with high precision, effectively automating manual data-entry processes and expediting proposal workflows
• Engineered an advanced hierarchical retrieval-augmented generation (RAG) framework integrated into a chatbot service, achieving superior semantic comprehension and significantly decreasing false-positive retrieval rates compared to industry-standard solutions
• Created automated algorithms for risk detection and compliance analysis in RFQ/RFP documents, utilizing recursive content checking aligned with organizational policies; results visualized in concise, interactive dashboards, substantially reducing stakeholder decision-making time
• Established foundational machine learning infrastructure and practices (MLOps, CI/CD pipelines, Dockerized deployments) from the ground up, enabling scalable, reliable production deployments and facilitating rapid experimentation and iteration cycles

**Siemens Experience: Machine Learning Engineer (April 2022 - September 2024)**

**Advanced Retrieval Augmented Generation for Industrial Copilot:**
• Designed and implemented complex RAG pipeline integrating HyDE, hierarchical retrieval, and iterative context refinement for optimized answer quality
• Developed context relevance and hallucination metrics, iteratively improving context selection and LLM prompting to enhance output accuracy and trustworthiness
• Built source attribution mechanism providing explainability and transparency in LLM generated answers
• Utilized state-of-the-art tools (OpenAI, Langchain, Haystack, PyTorch, HuggingFace Transformers) for implementation and evaluation

**Fully Unsupervised Semantic Document Segmentation:**
• Developed two novel unsupervised algorithms for semantic text segmentation, employing sliding window approach and sequential DBSCAN variant for improved context pruning in RAG pipelines
• Implemented algorithms using core scientific computing libraries (NumPy, SciPy, scikit-learn) and deep learning frameworks (PyTorch, HuggingFace)
• Filed invention disclosure currently awaiting patent approval

**Self-Supervised Knowledge Graph-Augmented Language Modeling:**
• Designed and implemented scalable multimodal encoder-decoder architecture for knowledge-graph augmented language modeling, utilizing PyTorch, Lightning, Hydra, and W&B for efficient development and experiment management
• Optimized graph embedding and subgraph retrieval techniques using FAISS and graph-tool, ensuring efficient knowledge integration and data processing
• Demonstrated structured research methodology through comprehensive literature review, model development, multi-task training, and rigorous evaluation on multi-GPU nodes

**Question Recommender System for Semantic Search Engine:**
• Developed context-aware query recommender system for QA system, enhancing user experience by generating relevant follow-up questions based on user behavior
• Integrated advanced table understanding techniques using fine-tuned models to focus question generation on parameter-related aspects, tailored for technical documentation search
• Implemented query recommender to optimize user workflows within parameter-focused context, particularly for simulation software documentation

**Early Career: Siemens Advanta Consulting (October 2020 - March 2022)**
• Proposed, researched, implemented and deployed production-grade, extractive semantic search engine utilizing retrieval and reranking techniques
• Implemented and evaluated various domain adaptation techniques (TSDAE, MLM, CT, SimCSE) for fine-tuning pre-trained models on proprietary data, leveraging cutting-edge transformer architectures for long document understanding
• Applied MLOps principles for efficient experiment tracking and results visualization
• Delivered scalable search solution using PyTorch, Hugging Face Transformers, and Sentence Transformers
• Built web application using FastAPI, Uvicorn, and Docker for seamless user interaction and production deployment
• Developed TypeScript with Angular frontend for stakeholder presentations

**Academic Background:**
• M.Sc. Computer Science, Ludwig Maximilian University Munich (2021-2024) - Focus: Self-Supervised Learning, Multimodal Deep Learning, Computer Vision, Knowledge Graphs, Deep Generative Models
• B.Sc. Computer Science, Ludwig Maximilian University Munich (2015-2020) - Focus: Machine Learning, Deep Learning & AI, Natural Language Understanding, Representation Learning

**Research Expertise:**
• Self-Supervised Knowledge Graph-Augmented Language Modeling (M.Sc. Thesis)
• Tutorial on Linear-Time Sequence Modeling with State Space Models (comprehensive Mamba tutorial)
• Parameter-Efficient Fine-Tuning for Vision Transformers
• Hierarchical 3D Attention Model for Video Understanding
• Self-Supervised Multimodal Alignment with Self-Distillation
• Co-authored book chapter on Taxonomy of Multimodal Deep Learning

**Technical Stack:**
• **ML Frameworks:** PyTorch, Lightning, TensorFlow, Hugging Face Transformers, Sentence Transformers, scikit-learn
• **MLOps & Deployment:** Docker, Kubernetes, FastAPI, Flask, Git, GitHub Actions, CI/CD Pipelines, AWS, Azure, GCP
• **Specialized Libraries:** Langchain, Haystack, FAISS, graph-tool, Hydra, NumPy, SciPy, Pandas
• **Experiment Tracking:** Weights & Biases, MLflow

**Available for:** Advanced ML Consulting, Multimodal AI Systems, RAG Implementation & Optimization, Knowledge Graph Integration, Production ML Pipeline Development, Research & Prototyping