<p>Conducted performance testing in a dozen different projects. Troubleshooted acute performance and stability issues in production. Prevented 18 performance incidents.</p>
<p>Design and development of a prototype disease management software in C#. Responsible for requirements, software architecture, and agile development. Sped up search by 1000x.</p>
<p>Conducted requirements engineering and feasibility analysis of <em>mm-accurate indoor localization</em> for a VR augmented welding plan for steel portal constructions.</p>
Arduino, GSM modules, load cells, undisclosed energy harvesting tech
</div>
<p>Evaluated technical feasibility of a smart drinking glass. Prototyped detail solutions for energy harvesting, weighing, and mobile communications.</p>
<p>Created an Android app to collect biosignals (PPG, ECG, BCG, SCG) and estimate heart disease risk. Implemented <em>signal quality estimates</em> and <em>event detection</em> from literature. Reverse engineered Kardia ECG encoding. Published arrhythmia detection, see "Publications". Validated use-cases at exhibitions and insurers.</p>
<!--
<ul>
<li>Designed and implemented an Android app to collect biosignals in a hospital.</li>
<li>Designed and implemented an Android app for heart disease risk screening.</li>
<li>Validated use-cases for B2C, including dozens of individual tests at several exhibitions.</li>
<li>Validated use-cases for B2B, including discussions with several insurance companies.</li>
<h3><spanclass="t">Research Associate, The University of Edinburgh</span><span></span class="d">(2015⁠–⁠2016)</span></h3>
<divclass="left-hang">
C++, Python, matplotlib, LaTeX
</div>
<p>Implemented domain adaptation and optimized runtime performance in a machine translation software. See "Publications".</p>
<divclass="left-hang">
mosesdecoder
</div>
<p>Implemented a mixture model as the language model to improve translation quality above that of Google Translate when used in a domain-specific scenario.</p>
<p>Sped up the translation model in an online-learning scenario by changing the data structure for the underlying suffix array, reaching an overall speedup of 3x.</p>
</div>
<divclass="block">
<h3><spanclass="t">Software Developer and Tester, Catalysts GmbH</span><span></span class="d">(2010⁠–⁠2014)</span></h3>
<divclass="left-hang">
C++, Java, Linux kernel, u-boot
</div>
<ul>
<li>Implemented SEO for a webapp for an online marketplace</li>
<li>Ported routines for processing satellite data</li>
<li>Organized tasks and infrastructure for a coding contest</li>
<li>Adapted bootloader and kernel of an embedded router for 4-byte flash memory</li>
<p>Designed a test device for connecting to and testing ticket printer hardware for buses, including <em>circuit design</em>, layouting, mechanical design and industrialisation.</p>
<p>Master thesis: "Handling out-of-vocabulary words in a domain adaptation setting in statistical machine translation"</p>
<p>Out-of-domain MT is sensitive to the alignment of rare words. 27 % of untranslated words in the output could be found in training data, but were too rare for the word alignment algorithm to extract.</p>
<p>Bachelor thesis: "Interfacing of the PARDISO Sparse Linear Solver for Schrödinger-Poisson simulation"</p>
<p>10-25 % overall speedup of simulations through refactoring the C++ codebase to introduce a representation of sparse matrices and interfacing a sparse solver package.</p>
Diploma project: "Distributed server monitoring"<br/>
Higher Institute of Technology (HTBLVA für EDV)</p>
</div>
<divclass="section-spacer"></div>
<h2>Publications</h2>
<divclass="block">
<h3><spanclass="t">MMT: New open source MT for the translation industry</span></h3>
<p>Bertoldi, N., Cattoni, R., Cettolo, M., Farajian, M. A., Federico, M., Caroselli, D., ... & Madl, D. (2017). The 20th Annual Conference of the European Association for Machine Translation (EAMT), 2017.</p>
</div>
<divclass="block">
<h3><spanclass="t">Smartphone-based paroxysmal atrial fibrillation monitoring with robust generalization</span></h3>
<p>Madl, T., & Madl, D. (2017). 31st Conference on Neural Information Processing Systems (NeurIPS ML4H 2017).</p>
</div>
<divclass="block">
<h3><spanclass="t">Automatisiertes Verfahren zur Dokumentation eines Verstoßes gegen die Straßenverordnung "Einfahrt verboten"</span></h3>
<p>(Automatic method for documenting offences against the traffic regulation "No entry")</p>
