Evgeny Tsvigun

Network traffic monitoring system on big scale, allowing its customers to define monitoring rules in Groovy, which is a main purpose programming language suitable for DSL creation. The system is implemented in Scala as an Apache Spark streaming job with stateful processing, able to accept rule code, code updates and parameters for defined rules. Its output is a stream containing risk events of interest (REI), including incoming network events IDs that triggered them and some explanation of underlying reasoning. Code and parameters updates are accepted in runtime, without interrupting the streaming job.

I acted as a lead developer of the project, my input was designing and implementing the initial POC version with Akka-Streams, designing contracts and interfaces for the incoming arbitrary code, designing and implementing the Spark based version, composing rule code examples library in Groovy, continuous requirement collection and filing tasks and requirements for project modules and accompanying tools. I am quite proud of my authorship of the initial concept and system design.

This is a Scala object, transforming a stream of updates of Groovy code for business rules defined by analysts into structures broadcasted to Spark worker. It intentionally doesn't contain business logic while showing somewhat tricky and non-typical Spark techniques usage.

I had a lucky opportunity to implement Hybrid Canopy-Fuzzy C-Means Clustering as a module of Apache Spark ML library. Following the linked research paper, I wrote it in Scala for Apache Spark with BLAS library for faster computations, strictly following the interfaces defined in Spark ML pipelines. That was a fun and enriching experience, and the results of applying the library to images were visually amazing.

The goal of the project was creating a scalable tool for a genetics lab, able to detect patterns of configured shape and density/sparsity in multiple genome sequences of related species of broad variety of living creatures, from bacteria to humans, and, naturally, the detected pattern's occurrences in input data.

Our starting point consisted of a white paper by Andrea Califano, describing potential approaches to the task.

We picked Erlang and self-written routines for cloud deployment as our tools. At the moment there were no ready distributed analytics engines to be found, so the main technical challenge was creating efficient task and data parallelization and task intercommunication.

The final result got us some well-deserved cheers from the client: it was able to take packs of human genomes ( quite long sequences compared to most bacterial genomes ) as input and had near-linear scalability, i.e. 2 EC2 instances produced the result ~1.95 times faster than 1, and the progression kept for bigger resources.

My role covered initial research, task planning, collaborative algorithm design, deploy routines and communication to the client. This particular project made me proud of my planning and communication skills.

It's a plugin for a well know Ruby-based task processing queue, Sidekiq. The plugin provides the ability to query job status and see the same in a web UI. It's an open source project distributed as a Ruby Gem, has some adoption. I initially coded it up and maintained it until I moved away from Ruby development, then I found worthy maintainers to continue the project. I'm happy to be the founder of an in-demand open source library.

It's a web resource for travelers, with some innovation on the part of geolocation functionality usage in Postgres DB ( PostGIS). This was my first lead position, where I provided technical guidance to a team of 5 developers and run our Scrum-based development process, collected and formalized requirements from the client. It was done with Ruby-on-Rails, PostgreSQL, PostGIS. The part that makes me proud, apart from successfully transforming initial design into a working product, may not even be essential - it's the interactive map-centric component with colorful buttons and location categories on the front page, produced by our graphic designer in collaboration with myself.

A framework simplifying on-chain game development. It makes use of

• State channels and highly formalized game rules definitions.
• EIP-712 signatures for off-chain game state updates.
• Session keys to allow users to skip sigining every move with their wallets.

GameJutsu is partially based on Magmo's ForceMove whitepaper, and it's aiming at being more developer-friendly and easier to use. ForceMove is a great protocol, it's all science-like and strict—i.e., it's hard to make a mistake and use it in the wrong way—but also it is harder to start using ForceMove for a less prepared programmer. GameJutsu is a bit more flexible and allows for more freedom in client-to-arbiter communication, and presumably requires less sophisticated code in game definition.

The project started as an award-winning entry at ETHOnline2022 submitted by a team of five.

Added vesting setup for multiple addresses per transaction, Ledger signing support to Bonfida vesting CLI tool. Also, I added a Python wrapper for scripted airdrops and coached the company stakeholders in using SPL and the new tools.