About

Hi, I'm Oleg Silkin! I'm currently working as a Machine Learning Engineer at Red Hat, where I work on the InstructLab/RHEL AI project. My primary focus is in the NLP space, mainly with training and evaluating LLMs. Most of my work thus far has revolved around developing the InstructLab Training Library to provide a consistent and reliable LLM training experience across a variety of systems which can range anywhere from low-resource machines to massive cloud-scale supercomputers.

Before becoming an ML Engineer, I worked on the Data Science team within Red Hat's Emerging Technologies department, where I helped develop and evaluate mechanisms for automating the process of generating backports for downstream repositories. This also helped me familiarize with many data science concepts and workflows, as well as tools such as Ray, OpenDataHub, and MLFlow.

Prior to this, I worked on the Red Hat Emerging Technologies Platforms team where I worked on various projects for automating & sustaining systems. Most notably, I worked on the copilot-ops project which served as a GPT-3 powered templating engine for an OpenShift/Kubernetes deployment. I also helped build the IPFS Cluster Operator, which was a collaboration effort between Red Hat and Protocol Labs to enable the streamlined operation of an IPFS Cluster inside of a Kubernetes Cluster (harder than it sounds!).

Due to my prior exposure to blockchain technology, I also ran a blockchain community of practice where I hosted various community members from the community to provide Red Hatters with exposure to the technology.

Before joining Red Hat as a full-time Software Engineer, I had the pleasure of interning on the Emerging Technologies Platforms team, where I was given much early exposure into building cutting-edge cloud-scale applications. A few projects which I was proud to work on are VolSync, a Kubernetes operator for data replication, and , an OpenShift environment which is optimized to run on single-node edge devices. During this internship, I also had the unique experience of working with Qiskit, IBM's quantum computing stack. Using this experience, I wrote a short blog on how to implement a 3-bit adder within a quantum computer.

Prior to my employment at Red Hat, I worked as an Open-Source Developer at LBRY.io, where I architected and implemented a full-stack social interaction system for a blockchain-based content platform, including a distributed comment service that integrated with blockchain validation, web frontend, and CLI interfaces. Additionally, I delivered various contributions to the Python-based SDK where I learned about how blockchain technology works. It was during this experience that I also learned about React.js programming and made numerous contributions to the LBRY desktop and Electron-based web app. This caused me to fall in love with user-facing applications, and full-stack development.

Besides coding, I enjoy learning about business, finance, and economics, and staying active to maintain a healthy work-life balance. Feel free to explore my portfolio and reach out to discuss projects or chat about tech and business.

Inspired by the ability of language models to parse contextual information, as well as frustration from the tedious job application process, I developed Plicanta.com as a single platform where a user can submit information about their career, and the application will attempt to generate an appropraite cover letter & resume with relevant information from the job listing. To build this, I had to integrate a number of different technologies such as S3 for uploading profile data, a PDF resume parser, Stripe integration for users to purchase in-app tokens.

In addition, I learned about all of the various features that startups often need to ship to deliver a product, and the complexity in maintaining them. For example, I implemented a free-tier which allowed all users to have 20 free generations which would replenish on a monthly basis. However; since users could also purchase credits, it was necssary to track these as separate fields on the backend but not emit this information on the frontend. This experience taught me that much complexity in delivering products lies not on the theoretical side but moreso on the product engineering side in getting all of these pieces to come together correctly. Of all of these challenges, the most logistically difficult by far was the implementation of a tutorial flow.

In the early days of ChatGPT, there was no notion of being able to upload and store documents that ChatGPT could readily read from and produce content from. For this purpose, I developed copywriter as an early-form RAG app for an email marketing agency to easily generate new marketing campaigns by being able to retrieve user data.

For fun I decided to re-implement Wordle in React Native to better understand how application development works. This was a super fun experience, since you get the feedback loop of working with React but with the satisfcation of having something running on your phone.

Framingham State University: Bachelor of Science in Math and Computer Science (2019-2022)

Harper College: Associate in Engineering Science (2016-2018)