Skip to content

README_ROADMAP

Jump to bottom
GN edited this page Jun 29, 2025 · 1 revision

NDF-Studio Roadmap

NDF-Studio is an open, mobile-friendly knowledge construction platform that uses Controlled Natural Language (CNL) to represent knowledge as node-neighborhoods. This roadmap documents the development milestones and architectural directions toward our first stable release.

✅ Current Milestone: Stable Backend + Single Page Frontend

  • Backend is stable using FastAPI + NetworkX.
  • Frontend is functional on both mobile and desktop.
  • Replaced Tailwind with UnoCSS for CSS utility support.
  • Monaco-based CNL editor fully functional.
  • YAML-based .ndf format is human-friendly and rendered from JSON internally.

🔄 In-Progress / Next Milestones

1. CNL Enhancements

  • Support qualifiers for nodes (e.g., "female mathematician") → Stored in node.qualifier
  • Support quantifiers for:
    • Attributes (e.g., "has size: at least 5 (cm)")
    • Relations (e.g., "most birds can fly")
  • Update the parser and CNL grammar accordingly.

2. Unique Node Identifiers

  • Implement unique or globally consistent node.id
  • Support document merging and deduplication using IDs.
  • Enables computation of scorecards across documents.

3. JSON as Canonical Parsed Format

  • Migrate parsed.yaml to backend-native parsed.json. parsed.json also removed. we use composed.json
  • Frontend renders JSON using 'composed.json` and composed.yaml used for NDFPreview.
  • Aligns with schema validation, testing, and frontend frameworks.

4. Smart Template Placement

  • Distinguish insertion triggers:
    • <...> → Relation
    • has ...: → Attribute
  • Improve template menu UX with semantic categorization.

5. Blockly-Enhanced Input (Optional UI)

  • Accordion-style block insertion panel above editor.
  • Basic users: Blockly blocks
  • Advanced users: Monaco text editor
  • Pre-existing Blockly block designs to be reused.

6. Scorecard (DevPanel → Statistics Tab)

  • Tentative scoring rubric:
    • Node: 1 point
    • Relation: 2 points
    • Attribute: 3 points
    • Node with qualifier: 2 points
    • Relation and Attribute with qualifier 3 points
    • Quantified relation/attribute: 5 points
    • Modality included for attributes and relations 5 points
    • Logical connective: 10 points
    • Argument composition: TBD
  • Rubric will be peer-reviewed and editable.

7. Modeling Processes

  • Backend support for storing multiple node-states
  • Add process_block in Blockly: transition from prior-state to post-state.
  • Prior-state and post-state are a set of node-states
  • Post-state of one process may serve as prior-state of another, like a petri-net
  • Backend schema updated to support role: process, 'state'

8. Lightweight Multiuser Support

  • Add support for user IDs without requiring email/mobile auth.
  • Store user workspace under /graph_data/users/{user_id}.

9. Accessibility Compliance

  • Follow accessibility best practices (WCAG).
  • UI and interaction audit to support learners with disabilities.

10. Localization Support

  • Internationalization-ready (i18n)
  • English-only in first release.
  • Language switching via user preference.

11. Help & Preferences as Graph

  • Implement Help and Preferences Tabs
  • Users can explore and edit their settings like knowledge units.
  • The Help document can be a readonly example graph document

12. Desktop Distribution (Electron)

  • Wrap frontend + backend in Electron for offline use.
  • Add Auto-update mechanism when online.

13. Testing

  • Unit tests for backend logic
  • Functionality tests for parser, graph storage, and UI
  • Test suite for CI-ready development.

14. Developer Documentation

  • Add docstrings in all Python and JS/TS modules
  • Internal dev documentation for:
    • Backend APIs
    • Graph schema
    • CNL syntax

✅ Completed Milestones (for reference)

  • Graph as document model (.ndf)
  • Freeform markdown + CNL per node section
  • Parsed graph rendering with Cytoscape
  • Monaco editor with save/parse hooks
  • Block templates for relation/attribute insertion
  • Soft Inverse relation generation, to distinguish between what is asserted and what is inferred (only in the frontend)
  • Graph templates and graph metadata
  • DisplayHTML card grid for NDF preview
  • UnoCSS styling for mobile/desktop layout
  • Delete graph, before the graph directory is removed, all mentions of the Nodes, and the graph in the nodes_registry.json to be removed, prior to deleting the nodes from user/nodes/ and then the graph directory from the user directory.

💡 Suggestions / Potential Enhancements

  • PEG grammar support for CNL parsing
  • Real-time CNL syntax highlighting
  • Plugin system for new CNL features
  • Interoperability with RDF / OWL via export adapters
  • Learning analytics and feedback suggestions

Version Milestone Tags

  • v0.1: Minimal working CNL + Graph + Blockly
  • v0.2: Scorecard + Process modeling + Electron App
  • v0.3: Accessibility + Localization + Help system
  • v1.0: Full offline-ready multiuser app with update support

🔍 Node Summarization & NLP Integration

  • Reintroduce automatic summary paragraph for each node.
  • Use spaCy-based NLP to extract:
    • Common/proper nouns → Node suggestions
    • Verbs/prepositions → Relation triggers
    • Adjective/noun phrases → Attribute suggestions
  • Parsed blocks rendered in DisplayHTML as:
    • Suggested Relations
    • Suggested Attributes
    • Suggested Node Types
  • Update DisplayHTML cards to show:
    • Node description
    • Extracted blocks (editable later)

🧠 Small Language Model (SLM) Exploration

  • Design or fine-tune a Small Language Model (SLM) for:
    • Generating single-paragraph summaries per node.
    • Operating fully offline or locally for cost-free inference.
  • Use open models (e.g., TinyLlama, DistilBERT) and fine-tune on:
    • Student-written node descriptions
    • Public knowledge bases (e.g., Wikipedia)
  • Output should be concise, factual, and paraphrased from CNL blocks.
  • Integrate SLM into generate summary workflow as fallback for missing descriptions.

Philosophy

“Structure should emerge from shared practice, not rigid formalism.”

License

To be confirmed. AGPLv3 recommended for educational openness.

Clone this wiki locally