Kaleidoscope

In a media landscape where most information is filtered, prioritized, and shaped by algorithms, Kaleidoscope explores what it feels like to encounter content without those invisible guardrails. The project is grounded in contemporary critiques of algorithmic mediation and draws on theoretical frameworks such as Bruno Latour’s Local–Global–Terrestrial model and research on political axes and media bias that question how knowledge is organized and categorized. By loosening those structures and removing personalization, Kaleidoscope treats search as a space for wandering rather than confirmation, inviting users to form their own connections across contradictory, overlapping, and messy pieces of information.

Technically, the system was built through a series of experiments in content curation, linguistic modeling, and randomness. We explored where randomness could be inserted into a pipeline (at the level of publishers, individual articles, or the pool of sources themselves) and tested how different placements reintroduced bias in subtle ways. We prototyped axis-based categorization using weighted word lists and preliminary NLP techniques (topic modeling, word-frequency scoring) to understand how articles might map onto ideological frameworks, but ultimately foregrounded randomness rather than classification. The backend pools multimedia from diverse APIs, scraped content, and manually collected sources, then assembles them without ranking: audio clips, tweets, images, citizen-journalism videos, headlines, and text fragments are all placed on equal footing.