Misha (Mikhail) Krasovitskiy, Senior/Lead Product Designer. Client product under NDA (2026): a graph explorer for public connections between people, organizations and events.
Image: Video: Skein graph explorer in motion
Skein
A visual graph interface that maps public relationships across figures, institutions, and events
- Role: Product Designer
- Project: NDA rebuild
- Year: 2026
- Platform: Web only
Context
The original product was built under a strict NDA for a private client. But the core problem was too interesting to leave out of my portfolio. So with the client’s permission I rebuilt the main experience: new brand, new data, same product thinking
The core concept: turn the classic detective’s evidence board into a digital workspace
Image: Skein: the same idea as a digital graph
Image: An evidence board with photos, pins and red string
Users
Journalists, analysts, researchers, detectives. Anyone who's job is to quickly understand how a person is connected to an organization or event, and verify every link back to its primary source
Stakes are high: wrong connection means bad data or even a wrong accusation
Constraints
High data density
Hundreds of people, organizations and events on one canvas
Data keeps changing
Roles and relationships shift very fast
Every fact needs a source
The data must be trustful
Legal risk
The product works with real people
Problems and decisions
PROBLEM 1
Splitting the screen between the graph and the data
I started by exploring how to balance the visual space between the graph itself and the context behind it. Internal A/B tests quickly pointed to the most practical layout: a dedicated left half for the graph and timeline and an independent scrollable right panel for the data
Image: Five layout options for splitting the screen between the graph and the inspector
PROBLEM 2
The graph turns into a hairball
Decision. Depth 2 by default. Node labels were restricted to primary hubs, revealing the rest on hover
PROBLEM 3
Exploring without losing context
Decision. Two modes: Preview on hover and Pinned on click. The panel rebuilds around the hovered node, while the pinned one stays the point of reference
Image: Video: Preview on hover, pinned on click
PROBLEM 4
Entity type is hard to read
Decision. Type is encoded by shape. Firstly, it was a special request by the compnay, since some of its workers were color-blind; secondly, because too many entities with their own colors would turn the graph into mess
Image: Legend: entity shapes and edge types
PROBLEM 5
Trusting a link
Decision. Every link has a date (Since) and a source. Past roles are dashed and tagged FORMER. Synthetic data is labeled SAMPLE DATA. No invented metrics about real people
Image: Department of Justice pinned, with a dashed former role
PROBLEM 6
One panel for diffrerent entity types
Decision. A shared template: header, main facts, connections, sources. But the set of facts depends on the type
Image: Right panel for a person
Person
Image: Right panel for an event
Event
PROBLEM 7
“Why is X connected to Y?”
Decision. The path is highlighted with a marker on the graph and explained in the panel (2 hops via Intelligence Community). The connections table tags it PATH
Image: Highlighted path from the pinned node to a hovered person
PROBLEM 8
Visual language
Decision. The challenge was balancing heavy data density with clean aesthetic hierarchy. Yellow became the main accent color because color-blind users could see it as well. And distinct node shapes were used to communicate data states beyond just color coding
Image: First concept
Before: first concept
Image: Final visual language
After
Stress test on real data
What broke once real data went in and how it was solved
OVERLAPS
More spacing plus a collision pass
Nodes and labels overlapped
MISSING PORTRAITS
Initials as a first-class state
Some people have no portraits. Initials are designed as a state
CHANGING ROLES
Role history and FORMER tags
Four cabinet members changed in 2026 alone
Outcomes
This was one of the most intellectually demanding yet deeply rewarding projects of my career, packed with pretty unique constraints. And my first time designing directly for color-blind accessibility. The live system was built for a specialized, closed audience using sensitive intelligence. Right now we’re actively gathering real feedback from users to shape the next iteration