Karlsruhe, Germany
M.Sc. Information Systems · Implementing AI, Data Science & Applied Research
I'm a Master's student in Information Systems at the Karlsruhe Institute of Technology who likes to build with AI — from LLM-powered applications and NLP pipelines to research on how reliably large language models actually behave. I pair that with a solid data-science foundation and consulting experience where technology meets finance and regulation. I enjoy turning messy, real-world problems into clear, well-engineered solutions — and I've lived and studied across five countries along the way.
01 — About
I'm Thanos Drossos — I grew up between Germany and the United States and have since studied and worked across Spain and Mexico. That international path shaped how I approach problems: curious, adaptable, and comfortable bridging disciplines — these days mostly around applied AI, data science, and information systems.
Lately I've been building and evaluating AI systems: conversational LLM applications, NLP and data-extraction pipelines, and current research at KIT on measuring how sensitive large language models are to the way prompts are phrased. Alongside that I bring hands-on consulting experience where technology meets finance and regulation. I care about rigorous analysis and shipping things that are genuinely useful.
02 — Experience
Dark Sky Data LLC
zeb
d-fine
finanzen.net
KIT · Institute of Applied Informatics & Formal Description Methods (AIFB)
Also active as a local buddy with the Erasmus Student Network (ESN) — supporting international students and organizing welcome weeks, events, and activities.
03 — Education
Karlsruhe Institute of Technology
Universidad Nacional Autónoma de México (UNAM)
Karlsruhe Institute of Technology
Universidad Politécnica de Madrid
Schuldorf Bergstraße & North Shore Country Day School
04 — Publications & research
ACM/SIGAPP SAC '25 · Catania, Italy
The paper measures the realised return of 30,058 closed Uniswap v3 liquidity positions across 9 pools over roughly 700 days, decomposing each outcome into impermanent loss and collected fees. The accompanying repository reproduces every figure in the paper from committed data.
@inproceedings{drossos2025amm,
author = {Drossos, Thanos and Kirste, Daniel and Kannengie{\ss}er, Niclas and Sunyaev, Ali},
title = {Automated Market Makers: Toward More Profitable Liquidity Provisioning Strategies},
booktitle = {Proceedings of the 40th ACM/SIGAPP Symposium on Applied Computing (SAC '25)},
year = {2025},
publisher = {ACM},
doi = {10.1145/3672608.3707833}
}05 — Projects
Developing Functional Information (FI_in), an information-theoretic measure of
how much a language model's answer changes when a question is paraphrased but its meaning
is not. Ongoing research at KIT.
Turns an archive of ~11,000 engineering drawings into something you can ask questions of. A vision LLM and a deterministic parser extract a structured record from each sheet, and a tool-using agent answers over the result while citing the drawings behind every claim. Extraction quality is scored against a gold set rather than asserted.
An agent over automotive crash telemetry in which every quantitative claim must cite the tool result it came from — enforced by an automatic citation checker and measured by an evaluation harness.
A wearable that detects nocturnal jaw clenching. Three electrodes across the forehead pick up the temporalis muscle, a per-user linear SVM classifies the frequency spectrum on an Arduino inside the sampling budget of a 1 kHz signal, and escalating feedback interrupts the episode before the teeth take damage. Trained per wearer, because EMG amplitude varies far too much between people for a shared model. KIT Ubiquitous Computing, WS 2025/26.
Tooling that automates compliance workflows for the EU Digital Operational Resilience Act (DORA), cutting manual effort in regulatory reporting.
More experiments and code live on GitHub.
06 — Skills & languages
Muted items indicate working / basic familiarity.
07 — Quick answers
08 — Contact
Open to working-student roles and thesis collaborations in AI and data science — and always up for a conversation where technology meets real-world problems. The fastest way to reach me is by email.