Alexander Tsvyashchenko

Google / Google AI team, Zürich, Switzerland

• Large-scale ML system research and development.

Daedalean AI / Detect and Avoid team, Zürich, Switzerland

• Research and development of the visual traffic detection system to detect uncooperative airborne obstacles:
  • Conducted a comprehensive survey on the available models and algorithms for ML-based object detection + tracking and led further experiments with the most promising approaches.
  • Led the research and development of multiple approaches for incorporating the temporal information into object detection ML models to increase the accuracy of the system.
  • Set up and integrated the Determined AI platform for distributed training, as well as optimized the training pipeline.
  • Designed and implemented the simulation of camera imaging pipeline to reduce the gap between the simulated and real data.
  • Co-designed and co-developed, with Simulation team, the full pipeline for specifying, modelling and generating the large volumes of the aircraft collision encounters.
  • Co-designed and co-developed, with Data team, the processes, data formats and tooling to ensure fully traceable data flows that are required for the system certification.
• Domains: Detect and Avoid, Machine Learning (ML Object Detection and Tracking), Visual Simulation, Cloud Computing.
• Skills: Tensorflow / Keras, Determined AI, Unigine, GCP, OpenCV.

Daedalean AI / Platforms team, Zürich, Switzerland

• Research and development of the production ML inference stack for flight-compatible hardware:
  • Gathered the input from stakeholders then researched, designed and implemented the full training-to-production models conversion and execution stack that targeted multiple deployment platforms.
  • Researched and implemented the proof-of-concept for ML kernels software verification to support both optimized execution and DO-178 certification goals.
  • Developed the proof-of-concept inference for ML models using FPGA-based Versatile Tensor Accelerator stack.
• Domains: Machine Learning + Inference, Compilers and Optimization, Certification and Formal Verification, Hardware (Embedded Devices, GPUs, FPGAs).
• Skills: Tensorflow, TVM, ONNX / ONNX Runtime, LLVM / Clang, Z3, Nix, Bazel, Yocto, Xilinx Vivado.

FAIRTIQ, Zürich, Switzerland

• Consulting FAIRTIQ’s research department on multiple ML-related questions.
• Skills: Tensorflow / Keras, sequence models, time series, features engineering, imbalanced datasets.

Google / Google AI team, Zürich, Switzerland

• Responsible for the full lifecycle of machine learning projects for multiple Google products:
  • Establishing and maintaining working relationships with the product team.
  • Scoping the problem and setting the goals with the relevant stakeholders.
  • Analysing the data and refining data collection processes.
  • Performing models research and experiments, evaluations and productionization.
  • Teaching product team members multiple aspects of ML in the process.
• Domains: ARCore (Tango), YouTube, Abuse, Ads.
• Skills: TensorFlow, TFX; multiple Google-internal technologies, languages and products; teaching ML classes inside and outside of Google.

Google / Ad Traffic Quality team, Zürich, Switzerland

• Traffic analysis to detect abusive / fraudulent patterns.
• Traffic quality signals, metrics and filters development.
• Tech lead for ad fraud botnets fighting efforts, including external information sharing and collaboration (e.g. BotConf-2014 presentation).
• Skills: C++, Python; multiple Google-internal technologies, languages and products.

Deebmedia, Amsterdam, Netherlands (remote)

• Responsible for the whole video processing and algorithmic stack: architecture and design, algorithms selection, implementation.
• In particular, R&D of robust, online algorithms for video background reconstruction in complex scenes and highly accurate object tracking.
• Skills: C++, Lua; multi-threading; boost, ITK, OpenCV, Qt; SWIG; Redmine, Subversion, CMake, Hudson, Simian, Vera++; Mac OS X.

Automated Industrial Machinery, Inc, Chicago, IL (remote)

End-to-end ownership of the following projects:

Atoms Optical Measuring, Locarno, Switzerland (remote)

• Consulted developers on numerical algorithms best suited for customer’s needs.
• Researched and implemented algorithm prototype for highly accurate image segmentation.
• The developed algorithm was used as one of the key components to achieve micron-level measurements accuracy.
• Skills: C++; OpenMP; ITK, OpenCV, WildMagic; Subversion, Trac.

Materialise, Kiev, Ukraine

• Researched and implemented complex, high-performance algorithms and data structures in computational geometry and rapid prototyping domains that formed the core of Materialise DigitalCAD kernel.
• Integrated and maintained them in Materialise libraries (≈ 1.5 million lines of code project), significantly improved libraries design.
• Guided research and managed research team of 6 researchers.
• Introduced multiple processes improvements in research team, including automatic testing introduction, autobuild system, code quality measurements, projects communication and documentation policies.
• Conducted trainings on Materialise libraries for the developers of Materialise products.
• Performed technical evaluation and made hiring decisions on candidates to the research team and other teams.
• Algorithms developed by research team served as the basis of principal Materialise products: Magics, Mimics, Simplant, 3-matic.
• Skills: C++; OpenMP, multi-threading; boost, CGAL, GTS, VTK; IncrediBuild; CruiseControl.NET, Subversion, VSS.

Multiple open-source projects

Developed multiple stand-alone projects and contributed patches to existing ones, some examples are below:

More information on projects is available here: https://endl.ch/projects

Kiev National University of Taras Shevchenko, Kiev, Ukraine

Cybernetics Faculty, Department for Theoretical Cybernetics