Applied AI engineer · researcher

Stuttgart, Germany
I’m an AI engineer based in Stuttgart. At additiv, I build multi-agent document workflows for regulated industries, particularly insurance and lending. My work takes these systems from early experiments into production—connecting the flexibility of language models with deterministic business rules, and keeping an audit trail of how each workflow reached its outcome. See my work
Outside work, I build tools around problems I’m curious about. ImmoPilot DE explores the numbers behind buying, financing, and renting out property in Germany. PDF Tools brings everyday document tasks into the browser without uploading files. Both are open source: ImmoPilot code and PDF Tools code. My interest in financial language also led to FinPlainLM, a dataset for training models to turn dense financial text into plain language.
Before additiv, I worked on machine learning security at Validaitor. There, I approached models from the other side: testing where they break, how they respond to adversarial inputs, and whether watermarking can help identify model copies. Alongside that work, my master’s thesis, KGGLDM, explored how knowledge graphs could guide diffusion models towards more controllable image generation.
I completed my M.Sc. in Computational Linguistics at the University of Stuttgart, where I worked across language, speech, and generative modelling. My research included how word meanings vary across time and geography, while SpeakerDiff explored generating speaker representations for voice anonymisation. I continue to share what I learn through technical writing and collections such as Awesome NLP Resources.
During that period, I also co-authored GlyphNet, which treats lookalike web-domain detection as a visual recognition problem. The work was accepted at the 2023 AAAI Workshop on Artificial Intelligence for Cyber Security, and the paper, code, and dataset are publicly available.
Before moving to Germany, I worked in India at Quantiphi, Scanta, and Mobile Programming. Those roles took me through machine translation, medical text analysis, and speech processing. I still develop an open-source speaker diarisation project, continuing that interest in understanding who spoke when.
My foundation was a B.Tech. in Computer Science from Dr. A.P.J. Abdul Kalam Technical University. My final-year project used deep learning to tag product catalogues; other student projects included School ERP, which remains on GitHub. Away from the keyboard, I enjoy table tennis, chess, cycling, and following financial markets.