Hello there, World! 👋

I'm Arijus Lengvenis, a passionate 4th year computer science student at the University of Durham, UK, currently on track to receive a high first-class degree. I have a a knack for creating impactful and user-centric solutions, low-level optimisation tasks to squeeze every tidbit of performance out a system, an AI enthusiast, avid coder, and constant learner. I've developed a handful of diverse projects, each providing me with a unique set of challenges that helped me grow both personally and professionally.

🧰 Languages and Technologies

• C++: Extensive experience (5 years) gained from high school through university on various projects, HPC, university assignments, coding competitions and more.
• Python: Some experience (3 years) gained through doing university assignments, AI programming, hackathons and various other projects, like automatic meter readings project.
• Webdev trio/ReactJS Some experience (3 years) gained through university assignments, personal projects, hackathons
• NodeJS/REST Some experience (3 years) gained through university assignments, personal projects, hackathons and internship at Wise.
• Java: Some experience from internship at Wise.
• C: Some experience from university courseworks.
• React Native: Some experience from university courseworks.
• Haskell: Some experience from university courseworks.
• Fortran: Some experience from ISC competition working with POT3D.

🎯 My Projects

🏰 The Royal Fork

My first project, the Royal Fork, is a feature-rich chess puzzle publishing and solving forum with a dynamic ELO rating system. It serves as an intuitive puzzle browsing and solving platform for chess enthusiasts of all skill levels. This project boosted my knowledge about web development tremendously as I ventured into the world of HTML, CSS, JavaScript, NodeJS, ExpressJS, RESTful APIs, Handlebars.js, user authentication, and unit testing using JEST.

Features:

• User-friendly puzzle browsing and solving platform.
• Fully-customizable puzzle editor with an ELO rating system.
• Interactive comment section for promoting user feedback and engagement.
• User authentication and tracking for personalized experience.
• Mobile and Desktop friendly user interface.
• Comprehensive Jest tests for frontend and backend functionality.

📚 Quizzy

Quizzy is an interactive web application for creating, organizing, and participating in online quiz competitions. It was here where I got my hands dirty with ReactJS, Websockets, OpenVidu, WebRTC and Google OAuth 2.0. Alongside PhrogLover, we designed and implemented a modern, interactive, and dynamic UI that encourages user participation and engagement.

Features:

• Real-time quiz sessions with voice, video, and text communication.
• Advanced quiz creation and editing tools for total flexibility.
• User profiles with comprehensive quiz and performance stats.
• Quiz lobby for communication, quiz controls, and a judging system.
• Authentication with Google OAuth 2.0 for seamless user login.

💬 IBM Chatbot for Cloud for Financial Services (C4FS)

This project was a significant step in my journey. I had the opportunity to work with a team of 4 on a self-learning chatbot to assist customers at IBM Cloud for Financial Services (C4FS). The project entailed using the IBM Watson machine learning framework for training a fully-functional chatbot to serve information from the IBM documentation promptly. I was the team leader, directly speaking with the client, leading Scrum meetings, and keeping things organized. The whole process taught me a great deal about cooperation within a team, reinforced my knowledge of agile development methodologies, and prepared me for my first summer internship at Wise.

Features:

• User-friendly interface for quick and easy query input and response retrieval.
• Integration with IBM Watson Discovery and Assistant for accurate results.
• A self-learning model that improves with user feedback.
• Auto-update function to keep the chatbot updated with the latest IBM documentation.

📱 LecTutor

LecTutor is a helpful active learning application that leverages the power of AI to improve the effectiveness of students' study time. The app automates the process of question creation based on lecture notes or slides, enabling users to study more efficiently.

Features:

• Question and Answer Generation: Using AI, the app creates a set of questions and answers based on the content of lecture slides or notes provided by the user.
• Review and Rating System: Users can rate the generated quizzes and review them later.
• Studying Reminders: The app sends reminders and incentives like a "daily streak" to help users keep consistent with their studying.
• Data Privacy: All user data is stored locally on the user's device, ensuring utmost privacy and security.

🚶 Dreamwalker

Dreamwalker is an advanced implementation of a Soft Actor-Critic model with Truncated Quantile Critics as well as Deep Dense Reinforcement learning, incorporating a Dreamer model to improve sample efficiency. The project focuses on a walking environment, specifically the bipedal walker and bipedal walker hardcore environments.

Features:

• Truncated Quantile Critics (TQC): A policy optimization method designed for continuous domains, aiming to reduce overestimation bias in Q-function approximation.
• Deep Dense Reinforcement Learning (D2RL): Enhances traditional MLP architectures by using dense connections to improve the policy and value functions.
• Emphasizing Recent Experience (ERE) Buffer: A replay buffer that gives priority to recent memories, thereby enhancing learning efficiency.
• Dreamer Model: A model-based approach using an auto-regressive transformer to increase sample efficiency of the agent.

Other Projects

• Face Generative Model: State of the art face generative model that is based on a VQGAN + absorbing diffusion design.
• Wise Recipients Page: Implemented a new design for the bank recipients page as well as added recipient creation functionality during my 10-week internship at Wise.
• Context-Aware Recommender System: Implemented a hybrid context-aware recommender system back-end, which utilises a personalised transformer and an embarrassingly shallow autoencoder to provide accurate movie recommendations.
• ISC 2023: During the ISC 2023 competition, was part of the ClusDur team and worked on running, optimising and automating the POT3D application for multiple CPU and GPU supercomputer nodes.
• Automatic Meter Readings: a computer vision based app that utilises a Arduino with a camera module to periodically take pictures of our power meter readings, use OCR to read the output and push readings to a Flask DB for storage, which is queried by a Discord bot.
• GymBattles: During Durhack 2020, worked on a computer vision application which tracked workout routine moves like pushups and integrated that into a fun competitive fitness app called GymBattles.
• Middle Ground: During Durhack 2021, worked on a debate app called Middle Ground, which offered users a voice chat-enabled debate playground to discuss controversial topics, with the goal of settling on a compromise between the two opinions, aka the middle ground, using AI.
• Stock Return Prediction: During Durhack 2022, worked on a stock price prediction model which utilised stock information from the last 20 days and tried to predict the stock price of the following day.

Among many other smaller projects.

🔨 Current Projects

• Automated HPL Benchmarking: I'm currently working on Python scripts to automate the High-Performance Linpack (HPL) benchmarking process. This includes setting up compilers, installing Message Passing (MP) and Linear Algebra (LA) libraries, conducting an optimal parameter search using Optuna, storing results in Prometheus, and visualising results in a dedicated dashboard. This work aims to streamline preparations for future Student Cluster Competitions (SCC) and allow participants to focus on more critical tasks.

• Parallel Euclidean TSP Solver: As part of my 4th-year project, I'm developing a parallel solver for the Euclidean Traveling Salesman Problem (TSP) for very large instances (n > 10^6). Leveraging my experience from SCC and my upcoming guest student programme at the Jülich Supercomputing Centre (JSC), I'm constructing a modular, parallel divide-and-conquer framework to handle large TSP datasets. The project is aimed to "help Santa deliver presents around the world", and would provide practical benefits for logistics and delivery company contexts.

👨‍💻 Coding Competitions

I enjoy participating in various coding competitions such as Leetcode and VarsityCode for fast, out-of-the-box thinking challenges. I also participate in Advent of Code to sharpen my problem-solving skills, algorithm and data structure knowledge, and learn new tips and tricks. Additionally, in 2023 I participated in the International Student Cluster (ISC) virtual competition, which helped me learn a lot about parallel programming, HPC and performance analysis. Our team ClusDur placed 5th in the interview round, where we displayed our results and findings.

🚀 Hobbies and Interests

When I'm not immersed in coding or studying, I enjoy:

• Playing tennis.
• Engaging in chess theory (and gameplay).
• Participating in quizzes (obviously).
• Constantly learning new things.

I truly enjoy engaging in brainstorming sessions and discussing intriguing ideas, even when they seem too ambitious for our abilities.