~/projects/crypto-arbitrage
Crypto Trading Arbitrage Platform
Posts by Collection
portfolio
~/projects/portfolio-3
nano-BERT
A simplified, from-scratch, comprehensible implementation of BERT inspired by ‘Attention Is All You Need,’ built to teach transformer fundamentals.
~/projects/portfolio-4
nanotorch
From-scratch reimplementation of core PyTorch internals: autograd, tensors, RoPE/attention, a pretrained tiny-GPT, KV cache, LoRA and full fine-tuning, and quantization.
~/projects/portfolio-5
Warehouse Management System
Full-stack warehouse management system: Django backend, Vue.js frontend, Kotlin Android app, and Raspberry Pi IoT integration for real-time package tracking and role-based access.
~/projects/portfolio-6
Space Model
Research-grade LLM framework improving zero-shot/few-shot text classification (HateXplain, IMDB, and others).
~/projects/portfolio-7
Crypto News Bot
Go microservice architecture using an LLM to summarize crypto news and auto-post to Telegram and Twitter, with multi-language support.
~/projects/portfolio-8
go-EdgeGPT
A CLI tool in Go for programmatic access to Bing Chat, built as an integration-ready open-source tool.
~/projects/stable-diffusion
LLM & Stable Diffusion Movie Generation Pipeline
publications
Ukrainian News Corpus As Text Classification Benchmark
Published in ICTERI Conference, 2022
In this paper we describe a framework for simple classification dataset creation with minimal labeling effort. We create a dataset for Ukrainian news classification and compare several pretrained models for Ukrainian language in different training settings.
Recommended citation: Panchenko, D., Maksymenko, D., Turuta, O., Luzan, M., Tytarenko, S., Turuta, O. (2022). Ukrainian News Corpus as Text Classification Benchmark. In: Ignatenko, O., et al. ICTERI 2021 Workshops. ICTERI 2021. Communications in Computer and Information Science, vol 1635. Springer, Cham. https://doi.org/10.1007/978-3-031-14841-5_37 http://stepantita.github.io/files/NewsClassificationBenchmark.pdf
Forecasting Bitcoin Price Trends: Leveraging Natural Language Processing and Bing GPT Data Augmentation for Enhanced Predictive Insights
Published in Unpublished, 2022
This study ex- plores the application of Natural Language Processing (NLP) techniques, specifically leveraging transformer models, to predict the impact of news articles on Bitcoin prices
Recommended citation: Tytarenko, S., & Lefebo, K. T. (2023). "Forecasting Bitcoin Price Trends: Leveraging Natural Language Processing and Bing GPT Data Augmentation for Enhanced Predictive Insights." Fordham Graduate School of Arts and Sciences, New York, USA. Email: [stepantita@fordham.edu, klefebo@fordham.edu]. http://stepantita.github.io/files/CryptoBERT.pdf
Breaking Free Transformer Models: Task-specific Context Attribution Promises Improved Generalizability Without Fine-tuning Pre-trained LLMs
Published in AAAI Responsible Language Model (ReLM) Workshop, 2024
Best Paper Award Spotlight Presentation AGI Leap Summit 2024 People's Choice — Fordham 3MT 2024
In this paper, we present a framework that allows for maintaining generalizability, and enhances the performance on the downstream task by utilizing task-specific context attribution
Recommended citation: Tytarenko, S. & Amin, M. R. (2024). "Breaking Free Transformer Models: Task-specific Context Attribution Promises Improved Generalizability Without Fine-tuning Pre-trained LLMs." AAAI Responsible Language Model (ReLM) Workshop, 2024. arXiv:2401.16638. http://stepantita.github.io/files/SpaceModel.pdf
Is context attribution all you need to attain generalizability in non-fine-tuned transformer? A Framework for Fake News Detection in Cross dataset Evaluation Settings
Published in ICLR Reliable and Responsible Foundation Models Workshop, 2024
We propose a novel method of context attribution for the transformer model that proves to be more efficient and generalizable. We show that in an example of a fake news detection task, utilizing three distinct datasets and outperforming the baseline model in both the same dataset and cross-dataset zero-shot test.
Recommended citation: TBD http://stepantita.github.io/files/CAM_Framework.pdf
Reflections on NeurIPS 2025: Advancing Evaluation and Continual Learning in AI
Published in Labelbox Blog, 2025
A Labelbox Research team reflection on NeurIPS 2025, covering rigorous evaluation and benchmarking of AI capabilities (data contamination, benchmark faithfulness) and the rise of agentic continual learning via reinforcement learning.
Recommended citation: Golchin, S., Modi, S., Tytarenko, S., Abdibayev, A., & Wetter, M. (2025). "Reflections on NeurIPS 2025: Advancing Evaluation and Continual Learning in AI." Labelbox Blog. https://labelbox.com/blog/reflections-on-neurips-2025-advancing-evaluation-and-continual-learning-in-ai/