OTHER_AI_SOLUTIONS

Metadata Generation

A minimal version is already implemented in PortalJS Cloud and can be used for demos:

Dataset creation starts with resource files instead of metadata
AI reads the files and pre-populates some metadata fields for the dataset and resources

What can be improved:

Dataset classification - The AI could read the list of tags and groups in the portal and suggest proper classification
Queryless-like data analysis flow - Current metadata generation can produce inaccurate results because the AI reads data files only partially. For example:
- A CSV with 10 thousand unsorted rows is uploaded, with columns such as "country", "date", and "value"
- AI reads the first 100 rows and tries to determine which metadata fields it can populate
- It guesses temporal coverage from the min/max values in "date" across those 100 rows
- It guesses spatial coverage from the values in "country" across those 100 rows
Schema awareness - The AI sometimes generates invalid metadata, for example a dataset name with special characters, which then causes a validation error in CKAN

This feature only makes sense for clients that create new datasets often.

Metadata Review

Not all metadata fields can be inferred automatically, and some still require human input, such as author or contact email.

After publishing, AI could review metadata to ensure it remains accurate over time. For example, spatial or temporal coverage may change in the data while remaining stale in the metadata.

For quick demos, this could be implemented as a skill in Queryless.

As with metadata generation, this only makes sense for clients that update the data catalog often.

Data Quality Audit

This is already partially implemented as a Queryless skill, with some standard data quality procedures:

Check if there are empty rows
Check if there are NULL or empty cells
Check if there are duplicate rows
Check if data types are consistent across the entire column
Check if cells contain values that look like anomalies, for example:
- If most values in the column range from "1" to "10", one cell with a value of "100" might be an anomaly
- If a column corresponds to a percentage, values over 100% might be an anomaly

The skill is enough to demonstrate AI-powered data quality audit, but the UX is not ideal.

Possible UX improvements:

Return a structured summary
Trigger the audit from a button or automatically on resource create/update
Send notifications to the dataset owner or portal admins
Display indicators in the UI

This only makes sense for clients that update data often.

AI-ready site optimization (GEO/AIO/AEO)

Optimizing content for AI and Large Language Models (LLMs) is generally referred to as GEO (Generative Engine Optimization) or AIO (Artificial Intelligence Optimization)

While traditional SEO focuses on getting a user to click a link, these AI-focused strategies focus on making your content clear and authoritative so AI platforms cite your brand as the primary answer

GEO (Generative Engine Optimization): The most common term for optimizing your brand and content so that AI search engines (like Perplexity, Google AI Overviews, or ChatGPT Search) summarize and cite your website.

AIO (Artificial Intelligence Optimization / AI Optimization): A broader term that covers optimizing your website and digital data so that any LLM (including models like Gemini or Claude) can easily understand, interpret, and recommend your services.

AEO (Answer Engine Optimization): Focuses specifically on providing direct, concise answers to AI algorithms, voice assistants, and featured snippets