July 10, 2026 · 5min read
How Feedzai’s Cosmos Explorer Reveals the Shape of Fraud
Do you ever look at the night sky and feel a bit overwhelmed by the sheer number of stars? That’s exactly how it feels to look at the billions of transactions flowing through financial systems every day. Each swipe, tap, or click is a tiny point of light in a massive digital universe.
Criminals take advantage of that vastness to stay hidden. They hope their shady dealings will get lost in the crowd. Feedzai’s Data Visualization Research team is currently building a “telescope” for this data; something we call Cosmos Explorer.
Key Takeaways
- Cosmos Explorer, an internal Feedzai research project, visualizes billions of transactions using an innovative “octree-based rendering” technique that transforms a massive, complex dataset into an observable environment for fraud discovery.
- Cosmos Explorer aims to dramatically improve speed to insight by using vibrant red highlights to make rare fraudulent transactions “pop” on the canvas, allowing analysts to quickly spot suspicious clusters.
- By combining familiar cosmic visual metaphors with natural language prompts, Cosmos Explorer offers an intuitive exploration platform to uncover complex, hidden fraud patterns that are invisible in traditional, flattened reports.
What is Cosmos Explorer?
At its heart, Cosmos Explorer, which is currently an internal research project, is a high-powered scatter plot that maps transactions by time and amount. But it’s much more than a basic graph. It uses “universe-related” visual metaphors to help humans process billions of events.
Think of it as a living map of transactional patterns. You can zoom out to see “galaxies” of known consumer activity, or zoom all the way in to inspect a single “star”, a lone transaction or a specific moment in time. It even uses natural language prompts, so an analyst can simply type, “Show me all hotel transactions over $200 from last week,” and the system highlights those specific stars in real-time.
How Does Cosmos Reveal the ‘Shape’ of Fraud?
In the world of finance, fraud is rare; usually representing less than 1% of all transactions. On a normal chart, that 1% is almost invisible. In Cosmos Explorer, we make fraud “pop” by changing the size and color of those specific points.
- Whitish tones: These shades represent the vast majority of legitimate transactions where consumer behavior follows predictable, “healthy” patterns. By using neutral tones for standard activity, the interface creates a calm background that prevents analysts from feeling overwhelmed by the sheer volume of data.
- Red highlights: We use vibrant red to mark fraudulent data points, ensuring they take immediate precedence over non-fraudulent activity on the canvas. These highlights act as visual sirens, guiding an analyst’s eye directly to anomalous clusters or “shady” sections within the digital universe.
RiskFM Foundation Model for Financial Risk
Financial services face an escalating arms race. Banks often reactively layer tools and data, resulting in disparate systems that offer a narrow view. Our whitepaper outlines why RiskFM, Feedzai’s an industry-first Universal Tabular Foundation Model, is the preferred approach to eliminate fraudsters’ favorite blind spots by ingesting data from multiple sources.
How Was Cosmos Explorer Conceived?
This project didn’t start with stars; it started with a “broken television.” Our first attempts to plot millions of points looked like a static-filled screen. We realized that even though we could render tens of millions of points, the human eye needs a better way to make sense of them all.
- Point density: We started with the idea of using brightness to show where transactions were most crowded, turning dense clusters into luminous clouds. This helped distinguish between heavy traffic areas and isolated events, though it initially lacked the clarity of our final version.
- Fish-eye lenses: We integrated a “details on demand” approach by using a fish-eye lens to magnify specific time windows. This allowed analysts to inspect the “nitty-gritty” of a single hour without ever losing the context of the larger day or week.
- Spectrograms: To differentiate types of activity, we experimented with machine-like representations that grouped transactions into broad swaths of color. These colorful bands acted like a spectrogram, helping us see how different merchant categories behaved before we fully transitioned to the cosmic metaphor.
How Was Cosmos Explorer Shaped by Our Understanding of Space?
The real “aha!” moment happened when we pinned a picture of the NGC 4565 spiral galaxy (about 30 to 50 million light-years away) above our desks.
We realized that transactions could be treated exactly like stars. We even sampled actual star textures to give our data points subtle variations in brightness and color. The result looks like a long-exposure astrophotograph, where bands of light represent dense activity and outliers appear as faint, wandering streaks.
To handle the technical side, we borrowed “octree-based rendering”, a technique used in the visualization of point clouds produced by 3D LiDAR scanning. This allows the system to break the data into smaller and smaller “cubes,” (like a Matryoshka or Russian nesting doll) showing a representative sample when you’re zoomed out and fetching finer details as you zoom in, allowing you to glide seamlessly through the data universe.
What are the Expected Benefits of Cosmos Explorer in Fraud Prevention?
Cosmos Explorer isn’t meant to replace the tools analysts use for split-second decisions. Instead, it’s a tool for discovery.
- Uncovering hidden structures: The tool reveals complex patterns that remain completely invisible in traditional tables, flattened reports, or more conventional charts aggregating data. It allows experts to see the “connect-the-dots” relationships that define sophisticated fraud rings.
- Speed to insight: By using immediate visual cues like red highlights and size changes, analysts can identify suspicious clusters in seconds rather than hours. This intuitive layout turns high-stakes data analysis into a fast, manageable exploration.
- Intuitive exploration: Cosmos Explorer makes complex financial data accessible by combining natural language (NL) prompts and familiar cosmic metaphors. It also integrates conventional insights charts and filter interactions (e.g., the cockpit of a spaceship) making it a highly intuitive and powerful tool for hunting fraud.
Historically, we have turned to the stars to seek patterns and link ideas amidst the dark. At present, we are utilizing that innate drive for discovery within the financial digital realm to expose obscured dangers. By transforming massive datasets into an observable environment, we leave no room for illicit actors to remain concealed.
Additional Resources
All expertise and insights are from human Feedzaians, but we may leverage AI to enhance phrasing or efficiency. Welcome to the future.