XAI- OGC API Processes for RAG/LLM using GeoDCAT and PROV

[rob-metalinkage]

The LangFlow toolkit uses Apache Airflow to define and run workflows using the LangChain toolkit. If such a workflow used or generated geospatial information sources, then wrapping it in OGC API Processess, and capturing details of the source, output and training data usage using the PROV model makes sense to integrate into spatial systems.

Potential codesprint outputs:

• GeoDCAT profile for PROV using LangChain workflow examples - extending and/or maturing the [existing Records-PROV Building Block - work in progress] (https://ogcincubator.github.io/geodcat-ogcapi-records/bblock/ogc.geo.geodcat.geodcat-records-prov) - note for a start this can simply demonstrate use of the generic PROV pattern supported with real workflow examples.
• Define a new Building Block PROV-AI profile for AI defining types of activities, using TrainingML
• GeoDCAT+PROV-AI profile (richer version of first with full description of activity and training set specifics
• Code for generic AirFlow components to capture provenance - wrappers for existing components?
• Code for LangFlow to capture specific provenance details
• OGCAPI-Processess profile for including PROV (draft to be available for review and testing)
• Code for AirFlow to generate OGC API Processess interface to export output and provenance
• STAC and OGC API Records extension/profiles for PROV-AI - building on STAC-PROV extension
• Examples and mappings for STAC as a profile of OGC-API Records mapped to DCAT - linking GeoDCAT-PROV to STAC-PROV and Records-PROV
• Code for AirFlow to generate STAC or Records metadata traces for generated objects
• Code for AirFlow to import STAC or Records metadata traces for referenced data objects and attach to the generated provenance trace
• extend existing Code for PyGeoAPI to deliver Records with PROV profile
• extend any OGC Records or STAC client to display provenance information
• extend and OGC Records or STAC editor to display and manage provenance information

These are all fairly small individually - but publishing the profiles as Building Blocks - or testing existing ones with these scenarios - ensures a output that can be built upon systematically, so any combination of the tasks has value to progress GeoDCAT and OGC APIs as an interoperability framework for XGeoAI - "Explainable AI for Geospatial".