DocViz turns any text — a paper, a report, a blog post — into an interactive map of its structure, meaning, and connections. Stop reading linearly. Start seeing.
A document is a graph wearing the disguise of a straight line. Its real shape — which claims support which, where the evidence lives, how concepts interlock — is invisible when you read top to bottom.
A 30-page paper costs an hour just to learn whether it's relevant. Most of that hour is navigation, not insight.
Arguments, citations and key concepts are buried in prose. You can't see the skeleton until you've read the skin.
Twenty documents become twenty silos. There's no shared map to see how a whole field — or your whole inbox — fits together.
Sections contain paragraphs. Paragraphs contain statements. Statements cite references and share concepts with one another. That hierarchy plus those cross-links is a graph.
DocViz extracts it losslessly — every statement preserved, nothing summarized away — then lets you fly through it at any altitude.
The same lossless logic runs locally or on the cloud worker — only the LLM's location changes.
Split into sections, paragraphs, statements & references — verbatim.
An LLM annotates each unit: one-liner, summary, key points, concepts, salience.
Assemble the hierarchy plus citation & concept cross-links into one network.
Surface co-occurring concepts, weighted by how often they bind paragraphs.
One control changes the altitude. Start with the shape of the argument; descend to the exact sentence and its sources.
The table of contents as a living map. The 10-second overview.
Each idea as a node, labelled by its own first sentence.
Every claim, sized by salience. The lossless core.
Citations as diamonds, linked to the claims that lean on them.
Plus a per-node detail slider: Skim → Summary → Analysis → Reference. Read as shallow or as deep as the moment needs.
The full living network
Proportions at a glance
Strict hierarchy
Radial brainstorm
Ideas, not structure
Lossless, searchable
Guided section-by-section
The model is open
An installable web app talks to Firebase; an autonomous worker on a GPU server does the thinking. Your laptop can be closed.
Installable, offline-capable. Drop a file, organize into folders, watch live progress. Network-first so every deploy lands instantly.
Auth, Firestore job queue & metadata, Storage for graph payloads. The rendezvous point — client and worker never talk directly.
A systemd service on the server polls for jobs, runs the pipeline on local Ollama, self-heals crashed jobs, and writes results back.
Today DocViz maps a document's content. Next, it reads a document's character — what kind of text it is, how good it is, and how it relates to everything else you've ever fed it.
An academic paper, a blog post and a legal contract are not read the same way. DocViz will auto-classify genre, then apply the analysis & scoring lens that fits. Tap a type:
Every node already carries structure. Layer scoring on top and a document becomes a profile — strengths, gaps and risks you can see in one shape, compare, and track over revisions.
When every document becomes a graph in the same space, your whole corpus becomes navigable. Concepts bridge papers; types cluster; quality colours the terrain.
One shared concept graph spanning your library. Find every claim about a topic, across every source, instantly.
Two papers side by side: where they agree, conflict, and cite the same evidence. Literature review as a diff.
Drop 200 papers in; get the terrain of a field — its hubs, its frontiers, its gaps worth filling.
The pitch: a fast-understanding layer for all text-based documents — read less, see more, decide faster.
DocViz began as a way to map a single paper. It's becoming an instrument for understanding text itself — its structure, its type, its quality, and its place in the bigger picture.