Data science · NLP · applied AI

Serhii Korsunenko

Senior Data Scientist and Applied AI Engineer. I build text-mining, NLP and LLM systems that are measured before they are trusted, and I take them from prototype to production. 15+ years, from critical infrastructure to Expedia-scale ML pipelines.

Open the live demo LinkedIn Email me
Featured research

Horizon Lens

Text mining on all 23,947 Horizon Europe projects from the EU's public CORDIS dataset. Screen a proposal abstract, search the portfolio with cited answers, follow research trends, detect overlapping grants, and compare classical NLP, transformer embeddings, LLMs and an LLM-maintained wiki on the same test set.

scikit-learnspaCysentence-transformers LLM + RAGLLM wikibias auditin-browser inference
0.58macro-F1 for TF-IDF, against 0.44 for the best LLM, at 0.02 ms per abstract
+0.13F1 on the smallest research fields after bias mitigation
388 → 16invented numbers caught and repaired by lint in the LLM wiki
5–0wiki answers vs RAG answers in an order-swapped judge, 3 ties
Temporal train/test split · human-validated gold set · paired bootstrap CIs Try Horizon Lens
How I can help

Text mining and NLP on your documents

Classification, entity and information extraction, topic and trend detection, duplicate and plagiarism detection, semantic search. Baselines first, so you know what an LLM actually adds.

LLM systems you can trust

RAG and knowledge bases with citations, schema-validated outputs, fallbacks, evaluation suites, and cost and latency budgets. Numbers come from code, not from the model.

Production ML and MLOps

Feature pipelines, model serving, regression tests and monitoring on AWS, Azure and Kubernetes, as built for high-availability systems at Expedia Group.

Evaluation and responsible AI

Temporal validation, A/B and experimental design, bias audits across groups and classes, and reports that decision-makers can read.

Experience
Nov 2025 – now

Independent Principal Applied AI & Data Engineer

  • LLM services with structured outputs, validation, fallbacks and evaluation
  • RAG and information retrieval connecting models to domain data
2020 – 2025

Senior Data Scientist & ML Engineer, EPAM Systems

  • Expedia Group: MLOps for high-availability feature pipelines and decision engines
  • ICA Sverige: predictive analytics and anomaly detection for a warehouse digital twin
  • Bayer: analytics and model validation on clinical data in a regulated setting
  • Azure OpenAI workflows that generate source-to-target data mapping documentation
2008 – 2020

Freelance AI & Data Engineer

  • NLP matching systems for freelance marketplaces, recommendation engines, ETL pipelines
2001 – 2007

Head of Computing Center, NPC Ukrenergo South Power System

  • Led teams running systems for a grid serving more than four million people
Toolbox

NLP & ML

  • Python, pandas, scikit-learn
  • spaCy, NLTK, Hugging Face
  • PyTorch, TensorFlow
  • Statistics, A/B testing

LLM engineering

  • RAG, agents, LLM wikis
  • OpenAI, Claude, DeepSeek, open models
  • vLLM, Ollama
  • Evaluation and guardrails

Platforms

  • AWS, Azure, GCP
  • Docker, Kubernetes, CI/CD
  • PostgreSQL, MongoDB, Elasticsearch
  • Databricks, Spark, FastAPI
Also

Writing

Books and practical guides on generative AI, LLM systems and applied machine learning, and articles on AI engineering and evaluation.

Products

I build and sell Inventory System, a SaaS for stock management, so I know what it takes to ship and run software for paying customers.