Christian Hammerschmidt Email & Phone Number

Delft, Zh, Nl

Us

I manage multiple teams of applied machine learning scientists, machine learning engineers, and MLOps/data engineers.Together, we are building, maintaining and improving abuse and fraud detection models at internet scale.

Us

Reporting to the director of ML, I lead and mentor applied scientists, and contribute to the applied ml research team. The team owns existing anti-bot, anti-fraud, and abuse products and develops novel solutions. I work cross-functionally with the product team, ML and backend engineering, and help develop the roadmap and ongoing prioritization of research efforts.

Den Haag, Südholland, Nl

APTA Technologies helps software-product owners develop a deep understanding of user behavior, identify suspicious events in log files, helping analysts and devsecops engineers to drill down to root causes of bottlenecks or breakdowns, and predict error or downtime occurrences, and monitor the impact on the business processes enabled by the software.Pando, our process analysis tool, extracts insights and knowledge from the application logs and visually presents the data to engineers, developers, and business process owners. Using these insights, root-cause analysis and predictive monitoring are set up.As the founder and technical lead, I took charge of greenfield application development, going from proof-of-concept prototyping to deployed production application. This included system design and architecture, CI/CD setups, end-to-end testing, and the MLOps lifecycle of the machine learning models powering our applications.Responsible for P&L and working cross-functionally, I interfaced with and lead team members in charge of UI/UX/front-end, backend, ML science, and took charge of product-market-fit validation and sales development, including customer-facing features such as interactive demos.

As a Software Developer:* In green-field settings, implemented and deployed ML pipelines as RESTful services using fastapi, deployed serverless on Digital Ocean and AWS.* Prototyped and deployed MVP web apps using a Python/Django stack, deployed serverless.As a Machine Learning Engineer:* Got rid of the need for pairwise training data for a Tesseract-based OCR service, leading to state-of-the-art OCR quality with minimal training data using a Pytorch-based CycleGAN deep network in preprocessing for image cleaning.* Implemented readily extendable performance evaluation pipelines of various preprocessing techniques over of standard classifiers and regressors using a variety of hyperparameter search algorithms, providing a framework for quickly evaluating novel imputation methods.As a Data Science Consultant:* Provided actionable insights into user behavior on a web app, quantifying the impact of new features on conversation rates and summarizing, visualizing, and writing up insights for non-technical users.

Delft, Nl

I develop machine learning tools for applications at the intersection of software engineering and security. I mostly develop novel algorithms and methods for anomaly detection, semi-supervised machine learning, and grammar inference. These techniques use online clustering, sketching, and locality sensitive hashing techniques in online settings, in part using Apache Kafka/Flink.Within the security setting, I also consider adversarial learning scenarios and was part of the winning team of the adversarial malware learning challenge at KDD.I mostly used C++, Python, and PyTorch. I also closely work and supervise students of all levels.

Replaced rule-based anomaly detection on components of transactions with an anomaly-based system, leading to increased accuracy and more flexibility in reaching AML/KYC compliance requirements for X number of users/transactions (using Python, scikit-learn, Pytorch).Built novel oversampling and imputation methods using deep neural network architectures such as GANs and VAE and supervised (PhD) students working on the topic (using Python, Pytorch, slurm cluster), leading to Y performance increase despite using simpler and more explainable models. THis led to improved imputation and oversampling techniques using deep learning.Acquired ~400k€ in funding.

I worked on sequence learning problems, in particular in extracting operational and generative models from data (in on- and offline settings). The models were used for (system) behavior prediction and anomaly detection in cyber-security. Initially working with Java/Spark, I later focused on Python/C++.I generalized state-merging algorithms to reverse-engineer variants of finite state machines. These models offer the possibility of integrating expert knowledge using (semi-/)supervised and active learning as well as inspecting these models further, e.g. comparisons with specifications, e.g. model checkers. The C++ implementation is available as open source.I showed that inferring high-level communication protocol descriptions from network traffic statistics offer great models for downstream tasks, resulting in state-of-the-art performance in fingerprinting and classification tasks.Additionaly, I worked on various projects with industrial partners, supervising master students working on probabilitic programming for data science applications (using Stan, Anglican) resulting in improved performance in automatic KYC/AML tasks.I supervised several interns and masters students.

Luxemburg, Lu

The Data Science Luxembourg hosts monthly talks on data science and machine learning across various application fields, reaching audiences in the BeNeLux area. Since mid-2018., the events are also live-streamed on Youtube. The meetup is currently sponsored by Luxembourg's SnT and Amazon.You can find us on Meetup.com, at https://www.meetup.com/LuxRgroup and https://datascience.lu

Chris primarily worked on a tech start-up that aimed to enable small businesses to provide customers with smart, personalized offers, and enable users to connect to their communities to find the best, most relevant values. In the process, he built prototypes using Cordova for mobile front-ends, used a mixture of python-eve and Django for the REST-API and web-frontend, as well as implemented the backend prototype using scikit's machine learning libraries in python.He also conducted interviews with a diverse range of potential customers and users, as well as partners for cooperation.With the offer to pursue a PhD in machine learning, Chris chose shift focus and not follow up on this project.

Berlin, Berlin, De

I researched the expressiveness of concurrent systems using process calculi and category theory constructions, capturing and describing properties of concurrent systems. The search group was particularly interested in building a hierarchy of expressiveness of various formalisms describing concurrent computing, e.g. through absolute separation results or establishing (semantic) equivalences.

Erlangen, Bavaria, De

As a software engineer, I made website editing accessible to everyone rather than technical admin staff by building a light-weight CMS transitioning the institute websites from static pages to an easily accessible templated-based WYSIWYG scheme.Implemented and maintained intranet functionality such as shared calendars, all built on a LAMP stack.As a teaching assistant, I consistently obtained top evaluations from students. In charge of sections for topics in theory of computing and category theory for graph rewriting.

Doktor (Ph.D.), Computer Science

Diplom, Informatik