Projects
My projects focus on AI-assisted research, reproducible data access, scientific software, machine learning, semantic technologies, and research infrastructure. I especially enjoy building systems that connect complex data sources, APIs, ontologies, and computational methods with researchers who need transparent and reproducible analytical workflows.
AI & cultural heritage
HeritageRAG — Evidence-grounded AI for cultural heritage research
An experimental retrieval-augmented generation system designed for cultural heritage research. HeritageRAG combines structured cultural heritage data, semantic vocabularies, retrieval methods, and large language models to produce research-oriented responses grounded in retrieved evidence.
The system integrates components such as Finna, Finto/YSO, and Annif for cultural heritage retrieval, semantic enrichment, and automated subject indexing. It also separates retrieved evidence from generated answers and preserves provenance so researchers can inspect the underlying sources.
Focus: retrieval-augmented generation · LLMs · semantic search · cultural heritage · provenance · R · Shiny · Quarto
AI-assisted reproducible analysis
A methodological research framework exploring how AI systems can help researchers construct transparent computational workflows from a research question.
The workflow covers data retrieval, filtering, cleaning, creation of analysis variables, statistical analysis, visualization, and reproducible reporting. AI-generated workflows can be evaluated against manually prepared reference analyses based on correctness, reproducibility, statistical appropriateness, and the amount of human correction required.
Focus: AI agents · reproducibility · statistical workflows · evaluation · provenance · research automation
Cultural heritage & bibliographic data
finna — Finna API for R
An open-source R package for retrieving and analyzing cultural heritage metadata from Finland’s Finna service. It provides reproducible programmatic access to records from libraries, museums, archives, and other cultural heritage organizations.
The package supports large-scale computational cultural heritage research by making metadata retrieval easier to integrate into reproducible R workflows.
Focus: API clients · metadata retrieval · reproducible workflows · digital humanities · cultural heritage
finto — Semantic vocabularies and ontologies for R
R tooling for accessing Finnish vocabulary and ontology services through the Finto infrastructure.
The project supports concept lookup, multilingual metadata enrichment, ontology-based exploration, semantic search, and the use of controlled vocabularies in computational research. It also provides an important semantic layer for projects such as HeritageRAG.
Focus: linked data · ontologies · multilingual metadata · semantic enrichment · APIs · R
annif — Automated subject indexing from R
R tooling for interacting with Annif, an open-source automated subject indexing and classification system.
The project connects automated subject indexing with reproducible R workflows and supports experimentation with semantic enrichment, controlled vocabularies, and AI-assisted metadata analysis.
Annif can be combined with Finto vocabularies and cultural heritage retrieval systems to provide automated semantic indexing as part of larger research pipelines.
Focus: automated subject indexing · NLP · controlled vocabularies · semantic enrichment · APIs · R
fennica — Finnish national bibliography workflows
Tools and computational workflows for working with Finnish national bibliography data at scale.
The project includes metadata retrieval, harmonization, enrichment, quality assessment, and computational analysis of historical publication data. It supports research into publishing history, authorship, language, genre, and long-term transformations in Finnish literary culture.
Focus: bibliographic data science · metadata harmonization · cultural history · large-scale data · R
Machine learning & geospatial data science
FinnishForestAI — AI and LiDAR for forest analysis
An applied geospatial data science project exploring how open Finnish laser-scanning and spatial datasets can be used for forest analysis with machine-learning and reproducible computational workflows.
The project demonstrates LiDAR point-cloud processing, extraction of forest structural features, spatial analysis, visualization, and preparation of geospatial data for machine-learning applications. Both R and Python workflows are explored, with an emphasis on open data and reproducibility.
Focus: LiDAR · remote sensing · geospatial data science · machine learning · forest analytics · R · Python
Graph learning for microbiome research
An exploratory machine-learning project investigating how microbial communities can be represented and analyzed as graphs.
Microbial taxa can be represented as nodes, inferred associations as edges, and abundance or functional information as node features. Graph-learning methods can then be explored for representation learning, prediction, community structure, and biological interpretation.
Focus: graph neural networks · microbiome · machine learning · network analysis · R · Python
Computational biology & scientific software
LimROTS
A hybrid statistical method and Bioconductor package for robust differential expression analysis.
LimROTS combines the empirical Bayes modeling framework of limma with reproducibility-optimized statistics from ROTS, supporting complex experimental designs and biological or technical covariates.
The method has been benchmarked using proteomics datasets and applied to real-world biomedical data.
Focus: statistical programming · proteomics · differential expression · Bioconductor · reproducibility · R
iSEEtree
An interactive explorer for hierarchical biological data, developed as part of a collaborative scientific software project.
The project provides interactive visualization and exploration capabilities for hierarchical structures in biological datasets and integrates with the broader Bioconductor ecosystem.
Focus: interactive analysis · visualization · hierarchical data · Bioconductor · R
Software & research engineering
Asynchronous data-processing APIs
A backend engineering project exploring scalable asynchronous processing using FastAPI, Celery, background workers, and REST APIs.
The architecture separates immediate API requests from longer-running analytical jobs, making it suitable for report generation, data processing, machine-learning workloads, and AI-assisted services.
Focus: FastAPI · Celery · Python · REST APIs · asynchronous processing · backend engineering
Open-source eCommerce platform
A modular eCommerce platform built around Drupal Commerce and modern development practices.
The project explores product and order management, payment integration, responsive frontend development, SEO-friendly routing, containerized development, automated deployment, and extensible commerce architecture.
Focus: Drupal Commerce · Docker · web development · payments · deployment · system architecture
Research direction
Computational and AI-assisted methods for cultural heritage research
My current research direction brings several of these projects together.
The broader goal is to investigate how data-access tools, semantic technologies, retrieval-augmented generation, and AI-assisted computational workflows can support cultural heritage researchers while preserving transparency, reproducibility, provenance, and human oversight.
Rather than treating AI as a replacement for researchers, the work investigates which stages of computational research can be reliably supported by AI and where expert judgement remains essential.
Focus: cultural heritage · RAG · AI-assisted research · reproducibility · semantic technologies · research methodology