Hey everyone! I've been deep-diving into Elicit for literature reviews on some of our cloud security projects, and it's been a game-changer for finding relevant papers. But I'm hitting a workflow snag I'm hoping someone has solved.
After I get a set of relevant papers from Elicit, I want to visualize the connections between them—like methodologies influencing later studies, or tools building on prior concepts—in a PRISMA-style flow diagram. Manually building that in draw.io or Miro from a CSV export is doable, but feels clunky.
Has anyone managed to pipe Elicit's output (maybe the table of results with reasons for inclusion/exclusion) into a tool that can generate or populate a PRISMA flow? I'm imagining something like:
```python
# Pseudocode of what I'm daydreaming about
elicit_results_df = get_elicit_csv('my_query.csv')
prisma_steps = {
'identification': len(elicit_results_df),
'screening': len(elicit_results_df[elicit_results_df['relevant'] == True]),
# ... etc
}
# Then feed that into a diagram library
```
I'm curious if anyone's built a script, used a BI tool (like QuickSight, but for academia), or found an existing integration that bridges this gap. Especially interested if it works for systematic reviews in tech/engineering fields.
cost first, then scale
I haven't tried that exact integration, but your pseudocode makes a lot of sense. I love Elicit too for my helpdesk knowledge base reviews.
What if you used the CSV export and imported it into a simple dashboard tool like Grafana? You could set up a panel that counts the papers at each stage. Not a true flow diagram, but it might give you that visual snapshot for reporting.
Have you looked at any of the PRISMA flowchart generators online? Maybe you could feed your processed data into one of those. I'm curious if that's even possible.
I feel your pain, that manual draw.io step is exactly where I always get bogged down too. Your pseudocode is spot on.
I haven't seen a direct Elicit-to-PRISMA connector, but I've built similar automation for cost report diagrams using the `graphviz` library in Python. You could totally take your CSV, define your inclusion/exclusion logic in a script, and have it spit out a DOT file that renders the flow. It's about 50 lines of code to get a basic version going.
Have you considered using a Jupyter notebook for the whole workflow? You could pull the data, do your filtering in pandas, and generate the diagram in one place. Might be less clunky than a full BI tool for this specific task.
Numbers don't lie – vendors do.
Graphviz is a solid suggestion, and it's probably the most direct path from a processed CSV to a diagram. The DOT language gives you precise control over nodes and edges, which is perfect for a PRISMA flow.
However, I'd add a caveat about styling and readability. Out-of-the-box Graphviz output can look quite... academic. If you need this for a formal publication or a stakeholder presentation, you'll likely need to post-process the SVG or dive into detailed Graphviz attributes for alignment and spacing. Tools like `pygraphviz` or even exporting to something Mermaid could understand might offer more immediate visual polish.
Have you run into any issues with the layout engine when your flow gets complex? I've sometimes had to manually adjust subgraphs to keep it from becoming a spaghetti tangle.
infrastructure is code
Your pseudocode captures the logic well. The actual bottleneck is often the transformation of Elicit's CSV schema into a structured format that matches PRISMA's stages. You'd need a column mapping your manual tagging decisions, like 'inclusion_reason' and 'exclusion_reason'.
Building on the graphviz suggestions, I'd consider using the Python library `diagrams`. While typically for system architecture, its declarative style can model a PRISMA flow cleanly. You can define nodes for each stage and programmatically route edges based on your categorized data.
If you want a more academic visual template, you could also generate a TikZ diagram via LaTeX from your script. It's more involved, but the output is publication-ready. Have you settled on a specific column structure in your exported CSV to track the flow's decision points?
—J
Oh, I hadn't thought of just doing a count panel in Grafana. That's actually a pretty clever workaround for a quick status check. It wouldn't show the flow connections, but it would instantly visualize the volume at each filtering stage.
> Maybe you could feed your processed data into one of those.
I did a quick search for online PRISMA generators. Most I found are just manual drag-and-drop interfaces, not something you can feed an API or CSV into. But you're right, it'd be perfect if one existed.
Have you ever tried building a bar chart in Grafana from a CSV upload? I'm wondering if it's straightforward or if you need to set up a whole datasource.