Buster.AI · AI / Enterprise SaaS · Media Intelligence
Buster.AI developed an AI fact-checking and media intelligence platform for news, government, and financial sector users.
Scope spanned the search platform, drilldown architecture, conversational AI search exploration, modular light-theme design system, responsive UI, and developer handoff. As sole Product Designer, I worked alongside the founder and lead ML engineers, building the architecture against a backend processing 2M+ articles daily across 50+ languages and global media sources.
The brief was lifecycle fact-checking. Tracking how a claim spreads, shifts in sentiment, and accumulates across sources over time.
Verification had to happen before publication, before investment, before action.
Drilldown was the verification mechanism. Different levels of complexity. Side-by-side data comparisons. The user picks the depth that the decision needs.
The engagement opened with discovery interviews with linguistic experts. Designing for AI fact-checking required understanding how language signals misinformation at the technology's underlying layer. The interviews surfaced what binary fact-checking cannot hold: most claims are slanted, distorted, or contextually inflected. Polarity holds the spectrum.
Five facets stacked under one query.
Polarity verdict. A five-point scale: Very Negative, Negative, Neutral, Positive, Very Positive, with sentiment score. The verdict layer where the AI's reading lands.
Lifecycle. Dual-axis box plot, line chart, and article volume bars on a single surface. The user reads the trajectory of a claim across the timeline of coverage.
Event clustering. Individual articles aggregate into events. Each event carries representative coverage, organisation, people, locations, and polarity verdict.
Source distribution. Geographic spread on a world map with bubble density. Top countries, source organisations, entity counts filterable across categories: Politics And Government, Executive Branch, Foreign Policy, and the rest of the taxonomy.
Article-level evidence. The deepest layer. Full article view with most-common-keywords treemap, taxonomy tree, source ranking. The point at which the user reaches the source claim.
Seven delivered components across the platform.
Three top-level modes: Search, Verification, Analysis.
Query input with autocomplete on entities, organisations, products. Filters across language (10+), source type, and polarity. Full filter panel covers location, entity, topic, interval, and content.
Sentiment over time chart with box plot and articles toggle. News volume area chart. Article cards tagged by polarity. Entities panel and geographic location view.
Article cluster aggregation with representative article, related coverage, and full event metadata. Polarity-tagged articles populate the cluster.
Natural-language query returns a structured, sentiment-backed answer. Four states: empty, input, thinking, answer.
Modular light-theme components built for multi-state coverage. Entities panel in 4 states. Sentiment box plot in 6+ states. World map at 4 zoom levels. Donut chart filterable. News volume area at multiple zoom levels.
Multi-breakpoint coverage of search and verification surfaces.
Conversational AI search, designed in early 2023. Natural-language query, sentiment-backed answer, four-state interaction.