[Feature Request] Automated Research Figure & Visualization Generation Pipeline

[koen666]

Problem

Currently, DeepGraph has no capability to generate publication-ready figures or visualizations. After running the SciForge validation loop, the system only produces text logs and numeric results—no charts, diagrams, or overview figures that could be used in a manuscript.

Evidence from Codebase

• Zero plotting libraries: No imports of matplotlib, seaborn, plotly, PIL, graphviz, or any equivalent across the entire codebase.
• No visualization Agent: There is no agents/visualization_agent.py or similar module.
• Frontend limited to navigation: web/static/js/app.js uses D3.js solely for a radial taxonomy tree (navigation UI). It does not render experiment result charts, method comparison plots, or knowledge graph subgraphs.
• Documentation/artifact gap: While HANDOFF.md mentions placing figures in artifacts/, there is zero code logic to actually generate those figures.

Missing Scenarios

1. Overview / Motivation Diagrams: The first figure in a paper (e.g., "Our Approach") showing existing method limitations vs. the proposed improvement.
2. Experimental Result Charts: After the validation loop, there is no automatic generation of bar/line charts comparing baseline vs proposed metrics.
3. Knowledge Graph Subgraph Visualization: The entity-relation graph exists in SQLite but cannot be rendered into a publication-quality graph figure.
4. Method Architecture Diagrams: No pipeline to translate a structured method description (from deep_insights.proposed_method) into an architecture diagram.

Impact

The SciForge closed loop currently ends with a text verdict (confirmed/refuted) and a final_report.md. Without figure generation, the system cannot produce a complete manuscript-ready artifact. A human researcher would still need to manually create all figures, which breaks the "autonomous scientist" vision.

Proposed Direction

Introduce a visualization_agent module with two tiers:

• Programmatic plots: Use matplotlib/seaborn to auto-generate experiment comparison charts, learning curves, and heatmaps from experiment_iterations and results tables.
• Conceptual diagrams: Use LLM-generated TikZ / Graphviz / Diagrams-as-code to produce overview and motivation figures from deep_insights content.

Figures should be saved into the experiment workspace (~/sciforge_runs/exp_N_*/figures/) and referenced in the generated final_report.md.

---

This is a structural gap that prevents the system from producing complete, publication-ready research outputs.