Projects
Machine Learning Engineer at GoodAIAI People Game
Nov 2023 - present
AI People is a sandbox game that leverages large language models (LLMs) to control behavior of non-player characters (NPC) and generate dynamic dialogue in real-time. As a ML Engineer, I developed and designed architecture of the game server API, fine-tuned small LLMs for local deployment, and orchestrated and managed multi-region deployments in Azure. I also led the development of the game distribution platform.
Large Language Models
LLM APIs
Python
FastAPI
PyTorch
JavaScript
React.js
Azure
Docker
Terraform
Stripe
Co-Founder & Tech LeadPortfolioMetrics
Sep 2023 - present
As a free time project, I developed and launched PortfolioMetrics, an online tool for backtesting and analyzing investment portfolios. The tool provides qualitative analysis to help investors make informed decisions, understand expected returns, and assess potential risks. Key features include performance analysis, risk assessment, portfolio growth analysis with cash flow strategies, Monte Carlo simulations, and Mean-Variance optimization, all integrated into a single, user-friendly interface.
Quantitative Finance
Python
FastAPI
JavaScript
Next.js
MaterialUI
Plotly
AWS
Redis
Docker
Terraform
Stripe
Lead Developer at PiVa AICheckFungi Mobile App
Dec 2022 - Oct 2023
CheckFungi is a mobile application that leverages AI to automatically identify fungi species from images, along with contextual information like location and time of observation. The application also enables users to request verification from expert mycologists. As the lead developer, I created the backend API and developed the web-based admin panel.
Computer Vision
Python
FastAPI
JavaScript
React.js
Docker
RabbitMQ
Postgres
Graduate Researcher at CTUFine-grained Visual Recognition with Side Information
Jan 2021 - Jan 2022
For my Master's thesis, I proposed a computer vision method for fine-grained visual recognition of snake and fungi species using state-of-the-art deep learning architectures (CNNs and Vision Transformers). To improve classification performance, I integrated contextual information such as location and time, adopted specialized loss functions to address class imbalance, and developed a weakly supervised localization method based on saliency maps. As a part of my work I participated in the SnakeCLEF 2021 challenge, where I applied my method and co-authored a working note paper.
Computer Vision
Python
PyTorch