• Deeper focus on GenAI/LLM Multi-Agent Systems (MAS) - organizing company R&D Task Forces - blog articles, white papers, webinars - building demo apps deployed on company-internal platform (terraform, AWS, GitLab) for sales pitches and marketing - further technologies (on top of tech in previous position): AutoGen, CrewAI, LangGraph, LangSmith, GPTs, Azure OpenAI Studio, Llama3, Qdrant• Public sector work on job skill extraction from job offers and CVs via classic NLP and LLMs- developed processing pipeline for Web UI in production (tens of thousands requests/day, <1s/document)- tools/technologies: Named Entity Recognition (NER), Named Entity Linking (NEL), spaCy, huggingface, transformer Modelle, MLFlow, hnsw, MongoDB, PostgresSQL, Bitbucket, Docker, pytest, poetry, Factory Design Patterns, pre-commit hooks, GDPR-compliant anonymization• Sales pitches• Interviewing

• Data Science and AI consulting• Webinars, seminars, workshops and hackathons about Data Science, Generative Ai (GenAI) and Prompt Engineering • Public sector and automotive industry• Development of LLM chatbots on custom data: - Retrieval Augmented Generation (RAG) via semantic search on Vector Databases (ChromaDB, Weaviate) - Use of various open and closed source LLMs and providers (OpenAI GPT-3.5, Meta Llama2, Databricks Dolly, NVIDIA Megatron, Aleph Alpha Luminous) - LLM Finetuning, Prompt Engineering, HuggingFace, LangChain• Development of AI image generator apps - comic generator app (streamlit, GPT-3.5, Stable Diffusion)• classic NLP: - Text classification, NER, SpaCy• Data anonymization and encryption for GDPR compliance• Streaming/telematics data analysis

• Combined vision neuroscience & data science | machine learning to dynamically predict neural activity to natural movie stimuli and study feedback loops and system robustness, with the goal to eventually emulate vision with a Deep Neural Network (DNNs) | Artificial Intelligence (AI)• 6 years of experience building Python-based SQL database pipelines with version control, code review, unit testing, performance optimization, remote access and integration with multiple collaborative lab code repositories • Custom-built, grid-search-optimized and cross-validated models, and classified neurons via regression, PCA, ICA, SVD, clustering and matrix factorization • Performed standard parametric and non-parametric, as well as custom-made bootstrapped statistical tests• Planned and executed complex long term project collaboration across three labs in Munich and Tübingen• Formulated scientific questions, developed, planned and conducted experiments, and collected, cleaned, analyzed and visualised data• Self-organized teamwork with fellow students for biweekly report deadlines to supervisors, using various project, task and time management tools for effective and adaptive schedules and priorities• Mentored and taught several MSc students, raised morale and motivation in the face of failures and got all of them onto publications• Collaborated with colleagues and contributed to their published projects• Scouted Master’s and PhD recruits to get best people into the team• Set up a lab wiki to iteratively consolidate our SOPs• Established a lab meeting agenda to capture task responsibilities, objectives and key results• Constructed a lab website to increase public outreach• Increased awareness of people management (regular student talks and lab survey to address potential issues)• Published in multiple peer-reviewed journals, gave talks and poster presentations at conferencesJoint supervisors: Prof. Busse (LMU), Prof. Berens (U Tübingen) and Prof. Euler (U Tübingen)

Graduate research in neural data science with my PhD advisors in the one-year programme offered by the German National Bernstein Network for Computational Neuroscience (BNCN), funded by the Volkswagen Foundation (30,000 €).www.smartstart-compneuro.de/www.nncn.de

Thesis title: 'In Vivo Optogenetics with the Inhibitory Step-Function Opsin SwiChR++ – Cortico-Thalamic Feedback Neurons for Robust Vision'Joint supervisors: Prof. Busse (LMU), Prof. Berens (U Tübingen) and Prof. Euler (U Tübingen)Description: Optogenetics has opened the gates for researchers to investigate the causal role of neuronal circuits with un- precedented spatio-temporal precision. However, achieving cell-specific, reliable long-term photo-inhibition in vivo remains challenging. We examined the Step-Waveform Inhibitory ChannelRhodopsin SwiChR++ (Berndt, Lee, Wietek, et al., 2015) in awake transgenic Ntsr1-Cre mice for long-term inhibition of layer 6 cortico-thalamic feed- back (L6 CT) neurons. Opto-inhibition was combined with visual stimulation and extracellular multi-channel electrophysiology in primary visual cortex (V1) and the thalamic dorsolateral geniculate nucleus (dLGN) to investigate the effects of SwiChR++ on spontaneous and visually evoked neural responses in those areas. Our findings indicate that this SwiChR++-mediated photo-inhibition increases neural firing rates in upper cortical layers while decreasing those in L6. Future analyses will quantify cell-type specific effects and determine the optimal optogenetics parameters. These preliminary results make SwiChR++ a promising tool to ultimately investigate what role cortico-thalamic feedback plays in shaping thalamic signal transformations.Methods: In vivo extracellular silicon-probe recordings from mouse early visual system, optogenetic manipulation of V1-activity, animal handling, surgery, microinjection of retrograde tracers/viral vectors, perfusion, histology, data analysiswww.cin.uni-tuebingen.de/

Supervisor: Marcel Oberländer, Computational Neuroanatomy GroupMethods: In vivo patch-clamp electrophysiology in somatosensory cortex of anaesthetized rats, biocytin-filling, perfusion, histology, and computational reconstruction of 3D-morphologywww.kyb.tuebingen.mpg.de

Supervisor: Thomas Euler, Retinal Signal Processing group• Simultaneous two-photon Ca2+-imaging and patch-clamp electrophysiology in ex vivo wholemount retinae of transgenic mice• Data analysis for spike inference using a probabilistic supervised machine learning model• Clustering of LGN-neuron types based on extracellular activity recordings

Thesis title: ’Expectations and Anticipatory Alpha-Modulations in Visual Distractor Inhibition’Supervisor: Prof Mark Stokes, Oxford Centre for Human Brain Activity (OHBA), Department of PsychiatryDescription: This project investigated how implicit experience of statistical regularities in the location of visual distractors (expectedness) improves the ability to ignore them, i.e. to inhibit their processing, and what role alpha-power-increases in V1 play in this type of inhibition. I carried out all aspects of the project, including experiment design, coding, human participant recruitment, data recording (behaviour, EEG), EEG time-frequency analysis and write-up.https://www.ohba.ox.ac.uk