Dmitriy’s Projects

Sample-efficient SQL fine-tuning

Sample-efficient LLM fine-tuning on a SQL database.

GPT-2 From Scratch to Tiny Stories

Building the GPT-2 transformer from scratch

arXiv LLMs Assistant

arXiv-llms-assistant is a ‘public’ version of a project submitted as part of an AI competition at LS Direct, demonstrating the use and evaluation of RAG with on-prem data as a way to capitalize on applications of gen AI to privacy-restricted domains.

Project Bird

GCP ETL project to pull data from eBird API 2.0 and display a map of recent bird observations in a given area - a small side project to practice GCP skills and see more birds.

Detecting anomalies in a statewide housing market with alternative data

Unsupervised learning (TDA) project to detect housing market anomalies built with Zach Stone for Haystacks.ai.

Investing in student housing with ML

Kaggle-style supervised learning

Inventory classes with pytest and pdoc

Inventory application for a tech video channel featuring computer builds with inventory classes in object-oriented Python, 90%+ coverage tests with pytest, comprehensive API documentation with pdoc, and packaging inside Docker.

COMPARATIVE STATICS OF A MONOPOLISTIC FIRM FACING RATE-OF-RETURN AND COMMAND-AND-CONTROL POLLUTION CONSTRAINTS

A paper I co-authored in mathematical economics/economic theory.