The technical side of Taras
Before the mountains and the coaching, there's a technical career: four-plus years building AI products, mostly LLM and NLP systems that go from idea to something a real business actually runs. If you need that side of me for your team, here's the work.
Experience
ML Engineer · Litslink
Sep 2023 – Present- Built an LLM-powered food-ordering chatbot (Flowise, Supabase, LangSmith), containerized and shipped with Docker and CI/CD.
- Trained a PyTorch company-classification model at 85% accuracy on a custom dataset collected via web scraping.
- Built a Slack AI assistant (FastAPI, MongoDB, Mistral API) that filters high-volume channels down to what actually matters.
NLP Engineer · HOLYWATER
Sep – Dec 2024- Developed a proof-of-concept RAG pipeline for semantic book evaluation using custom Hugging Face datasets and embedding-based ranking.
ML Engineer · MindCraft.ai
Mar – Aug 2023- Designed TensorFlow housing-price models: 9% MAPE for sellers, 3% MAPE for buyers.
- Built an end-to-end PostgreSQL data pipeline with automated cleaning and feature engineering.
ML / Deep Learning Researcher · NYU & Stony Brook University
Jun 2022 – Aug 2023- Responsible AI, NYU — studied how null-value imputation affects fairness and performance in ML decision systems.
- Stony Brook University — predicted 3D protein structure from amino acid sequence using PyTorch.
Projects
A few things built end to end, code included.
LLM Document Chatbot →
RAG-based QA chatbot for contextual document search and answers.
Geopolitical Data Analysis →
Public-dataset research testing hypotheses and visualizing trends in Python.
Ukrainian vs Russian Social Media →
Pure linear algebra and NLP — classified text origin with LSA at 93% accuracy.
Ukrainian Bible QA Chatbot →
Collected the full Ukrainian Bible text and built a QA chatbot on top of it.
Education & writing
Bachelor of IT and Business Analysis, Ukrainian Catholic University (2020–2024). Taught Python statistics seminars as an Econometrics TA. Thesis on vocal extraction in low-data regimes — paper, repo, demo.
Writes public notes on technical books: High Performance Python, Learning Bash Shell, Learning Unix.
Need an AI engineer, not a mountain guide?
Happy to talk through a project, a build, or a consulting engagement.