Drag & drop a PDF here
or click to browse
No graphs yet. Upload a PDF to get started.
Loading samples...
RPG — Research Process Graph
Transform scientific papers into interactive, structured knowledge graphs. Upload a PDF and let AI extract the research questions, methods, and findings as a navigable graph you can explore, validate, and share.
Quick Start
Go to the Upload tab and drag-and-drop a research paper (or click to browse). The system extracts the text and sends it to an LLM for graph generation.
Once processing completes you are taken to an interactive graph. Nodes represent research elements; edges show how they connect.
Search nodes, filter by type, chat with an AI about the paper, validate claims against the source text, or generate structured summaries.
Node Types
Every node in the graph is classified as one of three types:
The hypotheses or questions the paper sets out to answer.
Experimental techniques, datasets, tools, or analytical approaches used.
Results, conclusions, or observations reported in the paper.
Edges connect nodes to show relationships — for example, a Method that addresses a Research Question and produces a Finding.
Features
Upload any research paper as a PDF (up to 50 MB). The text is extracted automatically and an LLM generates the Research Process Graph. Progress is shown in real time.
Powered by Cytoscape.js. Pan, zoom, click on nodes for details, and switch between layouts:
- Hierarchical — top-down DAG (default)
- Breadthfirst — level-based tree layout
- Force-directed — physics-based spring layout
Open the chat panel while viewing a graph and ask natural-language questions about the paper. The AI has access to the extracted graph structure and (for uploaded papers) the full paper text. Responses stream in real time and support Markdown formatting.
Select one or more nodes and click Validate to check whether each claim is supported by the original paper text. Results are colour-coded:
- Supported — fully backed by the paper
- Partially supported — some evidence found
- Not supported — no matching evidence
Relevant excerpts from the paper are shown alongside each result.
Click Summarize to generate a structured overview of the entire paper. You can also select specific nodes first to get a targeted summary of just those elements. The summary streams in a modal and renders as Markdown.
JSON — download the raw graph data for programmatic use.
PDF — export a publication-ready snapshot of the graph with a colour-coded legend.
Use the search bar above the graph to find nodes by label. Toggle Q / M / F checkboxes to show or hide specific node types. Use the Select checkboxes to bulk-select all nodes of a type for validation or summarisation.
Explore pre-loaded sample graphs from the Samples tab to see how RPG works before uploading your own paper. All features except validation (which requires the original PDF text) are available on samples.
Tips & Shortcuts
| Action | How |
|---|---|
| Select a single node | Click on it |
| Multi-select nodes | Hold Shift and click additional nodes |
| Select all nodes of a type | Use the Select: All Q / M / F checkboxes |
| Clear selection | Click Clear selection button |
| Zoom | Scroll wheel or pinch gesture |
| Pan | Click and drag on the background |
| Reset view | Click the Fit button |
| Change layout | Use the layout dropdown (Hierarchical / Breadthfirst / Force-directed) |
| Open AI chat | Click the Chat tab on the right side of the graph |
| Send a chat message | Type and press Enter or click Send |
FAQ
What file formats are supported?
Currently only PDF files (up to 50 MB) are accepted. The system uses PyMuPDF to extract text from the PDF.
What LLM powers the extraction?
Graph extraction and chat are powered by OpenAI models. The extraction pipeline uses a more capable model for accuracy, while chat uses a faster model for responsiveness.
Can I edit the graph after extraction?
Manual graph editing is not currently supported. You can re-upload the paper to regenerate the graph.
Why is validation unavailable for sample graphs?
Validation compares graph nodes against the original paper text. Since sample graphs are pre-loaded without the source PDF, validation cannot be performed.
Is my data stored permanently?
Uploaded PDFs and generated graphs are stored on the server for the duration of the session. They are not shared with other users.