🌐 Community Open Source Roadmap

[simonprovost]

🌐 UrbanMapper Community – Open Source Roadmap

🧭 High-Level Themes

• Broaden model support (open-source LLMs, ML-assisted enrichers/imputers/filters).
• Deepen geospatial integrations (JupyterGIS, Auctus Search, SKRUB, scikit-longitudinal, LangRS).
• Expand data access & loaders (GeoJSON/KML/GeoPackage/DBs/PyDataverse/Overture).
• Strengthen visualization (3D, raster layers, PyAutark, MapLibre/Leafmap, Pretty Maps).
• Improve docs, tutorials, and community workflows (FAQs, video guides).

---

1) Pipeline Generator – Open Model Support

Goal: Allow seamless switching between GPT, Llama (Ollama), Mistral, Grok, and other OSS LLMs.
Action: Add new generators under pipeline_generator/generators.

Tip

Aim for provider-parity tests so contributors can validate behavior across model backends quickly.

• Add Llama (Ollama) generator.
• Add other OSS LLM providers (Mistral, Grok, etc.).
• Provide configuration for easy provider switching and parity tests.

---

2) JupyterGIS (by @geojupyter) Upgrade

Goal: Align with JupyterGIS ≥0.5.0 and improve UX.

• Upgrade UrbanMapper compatibility to JupyterGIS 0.5.0.
• Verify scroll behavior in identify panel.
• Verify saving analyses after annotation.
• If JupyterGIS ships color scheme utilities, remove custom equivalents.
• Add a user guide for all UrbanMapper x JupyterGIS integrations.

---

3) Spatio-Temporal Data via scikit-longitudinal or others

Goal: Support longitudinal datasets

Tip

Include a minimal example notebook that chains two enrichers into a LongitudinalDataset to accelerate onboarding.

• Integrate LongitudinalDataset to build spatio-temporal layers.
• Add examples showing multi-enricher workflows feeding longitudinal data.

---

4) LangRS as an Urban Layer

Goal: Evaluate and prototype LangRS (remote sensing image segmentation) as a new layer/mixin.

Important

Confirm licensing and API stability of LangRS before declaring the layer GA-ready.

• Engage with @MohanadDiab / Polimi lead noted in upstream issue.
• Provide an example in examples/Basics/2-Urban-Layer.ipynb demonstrating LangRS usage.

---

5) Imputer Module Expansion

Goal: Add richer GIS and ML-driven imputation strategies (inherit from abc_imputer.py).

• Implement literature-backed GIS imputers.
• Add ML-powered imputers.
• Prototype novel rule-based or Geo-LLM-powered imputers.

---

6) Filter Module Expansion

Goal: Broaden filtering beyond bounding-box drops (inherit from abc_filter.py).

• Implement literature-backed GIS filters.
• Add ML-powered filters.
• Prototype innovative/Geo-LLM-based filtering.

---

7) Loader Module – More Formats & Data Entrypoints

Goal: Support more geospatial formats and external data sources (inherit from abstract loader).

Caution

Be mindful of optional dependencies—gate heavyweight drivers (e.g., PostGIS/DuckDB) behind extras to keep base installs light.

• Add loaders: GeoJSON, KML/KMZ, GeoPackage, JSON, XML.
• Add DB-backed loaders: PostGIS, SQLite, DuckDB.
• Integrate PyDataverse for dataset access (HuggingFace already done via #45/#49 on UpStream).

---

8) Visualiser Module – New Visualisation Primitives

Goal: Expand visualization modes (inherit from abc_visualiser.py; align visual specs via discussion).

Tip

Start with thin adapters and shared style tokens so new visualisers stay consistent with existing UX.

• 3D renderer (e.g via PyAutark, Voxcity).
• Pretty Maps integration (aesthetic focus).
• Leafmap + MapLibre globe/3D.
• Explore additional innovative visualisers.

---

9) Urban Layer Coverage Expansion

Goal: Broaden feature coverage and new data sources.

• Add subways/tubes.
• Add railways.
• Add OvertureMaps-based layers (e.g., from_overture(type="road"\|"water"\|...)).

---

10) Documentation Upgrades

Video tutorials

• Identify top use cases.
• Record step-by-step video walkthroughs.
• Surface them in Getting Started > “Video Tutorials.”

FAQ

• Collect common questions (“What makes Urban Mapper different?”, etc.).
• Publish concise answers in docs/FAQ.

---

11) Brainstorming Tracks (Early Exploration)

• Machine Learning Module refactor from current proposed implementation on Upstream for flexibility.
• Optional MCP integration: https://github.com/MCP-Pipeline/mcpstack-urbanmapper (Model Context Protocols).
• Raster Urban Layer (upstream in progress) to merge into Community build.
• PyAutark visualiser (upstream in progress) as a new visualisation primitive.
• LLM-powered enrichers (design space exploration).

---

🤝 How to Contribute

• Comment on the issue with the item you’re taking.
• Propose design notes before implementation for complex items (visualisers, multi-dataset enrichers, new loaders).
• Follow contributor docs: https://urbanmapper-community.readthedocs.io/en/latest/CONTRIBUTING_DEV/
• Keep scope small; open follow-up PRs for iterative progress.