I'm looking for some advice on designing a visual research dashboard focused on entities rather than starting with traditional article searches. My goal is to kick off by typing in an entity name, like "Anthropic," and then see a graph displaying related entities such as OpenAI, Dario Amodei, and Google. From there, I want to click on any entity or relationship to view the underlying sources—like articles, papers, or filings—that support those connections. The idea is to understand the landscape first before diving into the documentation. I've already checked out several tools like Diffbot, Golden.com, Exa, and others, but they don't quite meet my vision. If you were to build your own version of this system, what would it look like and what features would you want to include?
3 Answers
I'd approach it with an entity-first mindset and a zoomable graph as the main feature. Add a side panel that updates with context like timelines, key players, filings, and sentiment trends. When you click on a node, show relationship strength and source density instead of just a boring list of links. It’s all about mapping first, then digging into the documents. If you’re looking to build it quickly, check out tools like Runable to scaffold your graph and panel layout; it’s not flawless, but works well for testing user experience before refining.
I don’t have much to add, but I’d probably utilize my framework-agnostic tiling window manager called ug-lay. It might not be better than others, but I think the name is cool!
Oh cool! I’ll take a look.
To create the kind of dashboard you're describing, think about a knowledge graph as the backbone with a retrieval layer. Start by using something like Neo4j for storing entities as nodes and their relationships as typed edges. When you type a name like "Anthropic," it should locate that node and pull up connected ones with a couple of hops. For extraction, run an NER model over your document pool—like SpaCy for basic needs or a fine-tuned LLM for more complicated relationships. Then, for visuals, consider a force-directed graph using D3.js or Sigma.js where node size reflects importance and edge thickness indicates how many sources support that relationship. Don't forget to factor in entity resolution to manage variations of names. Also, adding a temporal layer could be really valuable to observe how relationships change over time! Plus, using something like Diffbot's API might help get a quick prototype together.
Wow this is gold! I’m currently replying on my phone. I’ll properly read it again tomorrow. You sound like you know this space well.
Amazing!! Really helpful. I like the increasing navigation part!