Candid Garden
Candid Garden is a research-driven digital platform for exploring art and music through semantic search and AI-assisted analysis. It is designed as an experimental environment for studying how multimodal AI systems can support large-scale cultural analysis, interpretation, and discovery.
The platform implements vector-based semantic search across a growing corpus of over 54,000 artworks and 2,100 curated music tracks. Rather than relying on exact keyword matching, Candid Garden models semantic relationships between images, texts, and concepts, enabling concept-driven queries that operate across themes, emotions, iconographic motifs, and abstract ideas.
A core research focus of Candid Garden is the development and evaluation of AI-assisted tagging systems. Artworks are enriched using structured metadata derived from both human-generated tags and AI-generated annotations. These annotations are organized according to art-historical and semantic frameworks, enabling comparative analysis between human and machine interpretations at scale.
In addition to search, the platform includes experimental recommender and exploration features that expose users to related artworks and musical pieces based on semantic proximity rather than popularity or predefined categories. These features are treated as research instruments, allowing the study of how AI-mediated recommendations shape discovery paths, attention, and interpretive practices.
Candid Garden is developed as an open, inspectable system with the aim of increasing transparency around how semantic search and recommender systems operate in cultural and research contexts. The project prioritizes reproducible methods, structured data, and empirical evaluation over opaque or purely aesthetic AI applications.
About the Researcher
I am Osman, a researcher and web application developer working at the intersection of AI, digital humanities, and the social sciences.
I am based in Munich, Germany. I hold a Bachelor’s degree in Economics from Yildiz Technical University (Istanbul) and a Master’s degree in Economics from Ludwig-Maximilians-Universität Munich (LMU), with a strong focus on quantitative methods and empirical analysis.
I have been involved in a research project collaboration at LMU with Professor Kohle, situated at the intersection of art history, digital methods, and cultural data. Within this context, I worked on the automation and structuring of art-historical metadata using crowdsourced datasets from the ARTigo project.
The research focused on training AI systems to recognize not only pre-iconographic elements—such as objects, figures, or settings—but also iconographic and iconological concepts in representational paintings. This enabled complex semantic queries, for example exploring how religious figures, social status, or symbolic themes are represented across historical periods.
From 2018 onwards, I experimented with early computer vision and language models and later transitioned into industry, where I gained experience building and maintaining production-scale AI systems. The rapid development of multimodal models, including CLIP and large language models, made it possible to revisit and substantially advance the original research questions.
Candid Garden emerged from this trajectory as a research platform that integrates contemporary AI methods with empirical, theory-driven inquiry. It reflects my broader research interest in human–AI hybridity, semantic systems, and the ways AI reshapes knowledge work, interpretation, and discovery.
I view Candid Garden as an evolving research environment rather than a finished product—one that supports experimentation, evaluation, and open discussion around the design and consequences of AI-assisted semantic tools in cultural and institutional contexts.
For academic inquiries or collaboration, I can be reached at:
📬 hey[at]candidgarden.com