Projects

Fine-tuning a Finnish BERT model for domain-specific NER using CSC LUMI A100 nodes. Full DDP training, Weights & Biases logging.

Apache Kafka + Spark Streaming pipeline ingesting IoT sensor data from the KAMK campus building, anomaly detection with Isolation Forest.

Benchmarking open-source LLMs (Mistral, LLaMA, Phi) on Finnish language tasks. Results published as open dataset.

A containerized, self-hosted MLOps stack using MLflow, DVC, and Prefect — deployable on any Linux server or CSC cPouta.

Time-series forecasting of Finnish electricity demand using Temporal Fusion Transformers. Data from Fingrid open API.

Ready-to-use Jupyter environment configs for CSC Notebooks — pre-installed with common data science and ML libraries.