Software Engineer
Loading posts...
Feel free to contact me at kanishksachdev@gmail.com
The complete English Wikipedia link graph, live in your browser: 7.2 million articles, 707 million links. Search an article, expand what it links to, race between topics, or render the whole map at once.
Drop an article on the canvas, then double-click it to unfold the pages it links to.
Every blue link on Wikipedia is an edge in one enormous directed graph. I built that graph once before as an offline analysis; this is the live version, an interactive Wikipedia graph visualization backed by a small Rust service that answers lookups, link expansions, and shortest-path queries over all 707 million links in milliseconds. Each mode in the switcher is a different lens on the same data.
The default mode is a force-directed graph explorer. Drop any article on the canvas and it arrives with its closest neighborhood already connected; double-click a node to expand the pages it links to, colored by the community each article belongs to. The canvas fetches every real link between the articles on screen, so what you see is the actual network, not a tree. A shortest-path tool plays the classic six degrees of Wikipedia game exactly: pick any two topics and it returns the minimal chain of links between them, or the full web of every shortest path at once. The reckless can load an article's complete link set with no cap, all 1.66 million in-links of ISBN if they insist.
Race is the wiki game played properly: get from one article to another by clicking real outgoing links only. After every hop the server reports how many clicks you are from the target, a par tells you what the optimal chain would take, and giving up reveals that chain drawn on the map. Three difficulties, shareable results, and it works one-handed on a phone.
Where Explore is curated, Galaxy is exhaustive. It renders an article's complete neighbor set as WebGL points on a rank-ordered spiral, hundreds of thousands of articles when the topic is big, with every link among those neighbors drawn as a glowing mesh. One click runs community detection on that neighborhood in your browser and recolors it by the sub-topics it actually contains.
Atlas draws all 7.2 million articles as a single point cloud, laid out so that Wikipedia's communities form continents with names: the books continent around ISBN, the places continent around Demonym, biology, film, sport. Zoom out and community names label the map like countries; zoom in and individual article names take over. Firmament layers the link structure itself on top, pre-rendered tiles of all 707 million edges glowing like filaments, with the strongest links drawn live as you dive in.
Lab is Wikipedia network analysis in an editor: write a small JavaScript function against the graph's typed arrays (positions, communities, PageRank, edges, BFS distances) and the renderer draws your per-node colors and sizes over the map. Presets cover degree coloring, click-distance from any seed article, and label propagation; scripts run sandboxed in your browser and are shareable by URL.
The backend parses a full Wikipedia dump into a compact adjacency index plus a PageRank score per article, then serves it from a memory-mapped file in Rust, so queries touch the page cache instead of a database. Shortest paths run as a bidirectional BFS over the complete graph, which is why the six degrees between two obscure topics resolve about as fast as one hop. Bulk structures (complete neighbor sets, the whole-graph atlas, edge skeletons) ship as packed binary typed arrays that the browser renders directly in WebGL2, and the front of it all is a Next.js page with a d3-force canvas. The pure graph logic lives in tested, DOM-free modules.