Hacker

Divyansh Khurana

@divyanshkhurana06

Joined Oct 2026Student at VIT VelloreIndia

Elsewhere

Projects

  • Top 50

    Seismo

    Graph HacksOct 2026

    News doesn't get written on Wikipedia. It gets edited. When something happens in the world, strangers who have never coordinated converge on the same handful of articles within minutes, often in several languages. That convergence has a shape, and the shape appears before any headline does. Seismo is a newsroom whose only sense organ is that shape. It streams every human edit on every Wikipedia (about two per second) into FalkorDB as a live graph of pages, editors and edits. Every 45 seconds, pure Cypher finds pages where unusually many distinct humans just converged, builds a projection on the fly, and uses FalkorDB's WCC and PageRank to turn hot pages into story components with an epicenter. No LLM is involved in detection. A crew of Claude agents then acts on the graph. The Scout checks graph memory first (is this cluster already a Story, is a Task open?) and only then decides new / update / ignore. An Analyst claims the Task with one Cypher mutation, reads the dossier from the graph, pulls the real diffs from Wikipedia, and files a Brief where every claim has a CITES edge to the exact revision it rests on. A Desk agent answers questions with clickable citations. Restart the process and the crew remembers everything, because its memory is nodes in the same graph. Who it's for: newsrooms, OSINT analysts, trust and safety teams, and anyone who wants to know what the world is paying attention to right now, with evidence they can click. In its first minutes live it filed a cited brief on an explosion at a Canadian military base while correctly dismissing a vandalism spree as "not news". Built solo during the event. An AI coding assistant (Claude) was used; the idea, graph model, detection design and all decisions are mine. Anonymous editors are hashed before storage; no IP addresses are kept.