Instruments.

Tools change; the method doesn’t. This is the current bench.

AI / Machine Learning

We build and evaluate models where off-the-shelf ones stop working: large language models (OpenAI, LLaMA) — fine-tuning, RAG pipelines, generation quality evaluation; classical ML with scikit-learn; text-to-image (Stable Diffusion); GPU work with CUDA.

Data & Infrastructure

Research is reproducible only when the data layer is disciplined: Python (Pandas, NumPy, PySpark), ClickHouse, BigQuery, Postgres, Hadoop/Hive, orchestration with Airflow and dbt, packaging in Docker/Kubernetes, cloud on GCP.

Analytics & Processes

Finding the mechanism behind the numbers: process mining (Pm4Py, Disco), A/B experiment design, BI and monitoring (Power BI, Tableau, Grafana).

Have a problem no one has solved?