Skip to main content

Generative AI Services

Generative AI services leverage Large Language Models (LLMs) to perform complex tasks like intelligent cleaning, semantic reasoning, and automated data enrichment through natural language prompts.

Custom (LLM Reconciler)

A flexible LLM-powered reconciliation service that matches text values to entities based on a custom prompt. Use the LLM to intelligently reconcile data to knowledge base entities with custom matching logic.

LLM model: phi4-mini

Input: Column values to reconcile and custom instructions for matching.
Output: Entity matches with IDs, labels, descriptions, types, and confidence scores.

How to use: Write instructions describing how to match the values. The LLM will receive each cell value and return entity information.

Example: Prompt: Match this location to a Wikidata entity. Return the entity ID, name, description, and confidence score (0-100).

Custom Wikidata (LLM Reconciler)

A flexible LLM-powered reconciliation service that matches text values to Wikidata entities based on a custom prompt. Use the LLM to intelligently reconcile data to Wikidata knowledge base entities with custom matching logic.

LLM model: phi4-mini

Input: Column values to reconcile and custom instructions for matching.
Output: Entity matches with Wikidata IDs, labels, descriptions, types, and confidence scores.

How to use: Write instructions describing how to match the values. The LLM will receive each cell value and return Wikidata entity information.

Example: Prompt: Match this location to a Wikidata entity. Return the entity ID, name, description, and confidence score (0-100).

CH Matching - Private

An LLM-based Open Opportunities company house matching service. It uses a specialized hybrid search over a collection of European company data plus an LLM to reason about ambiguous or partial matches.

Input: A column containing company name values (can be reconciled names or raw text).
Output: New columns such as:

  • company_official_name
  • company_number
  • company_address
  • Additional columns with LLM-based reasoning explanations (e.g., llm_match_reason, llm_confidence) that describe why a match was selected and any normalization or assumptions applied

COFOG (LLM Classifier)

A classification service that assigns a government department or public organization to the most appropriate category among the ten top-level COFOG (Classification of the Functions of Government - For more details, click here.) options. The classification is based on the organization's name, description, country, and Wikidata metadata.

LLM model: phi4-mini

Input: Organization details including name, description, country and optional Wikidata fields such as description or type.
Output: New columns containing:

  • Predicted COFOG category (cofog_label: 01-10).
  • Confidence level (high/medium/low).
  • Reasoning notes

Custom (LLM Extender)

A flexible LLM-powered extension service that generates new columns based on a custom prompt. Use the LLM to intelligently enrich your data by creating additional attributes derived from existing column values.

LLM model: phi4-mini

Input: Column values to extend and custom instructions describing how to extend them.
Output: New columns populated with LLM-generated content based on your specifications.

How to use:

  • Specify output column names (comma-separated, e.g., 'summary, category, sentiment').
  • Write instructions for what to do with the cell data. The LLM will receive both your instructions and the actual cell data.
  • For each column, specify what value should be returned.

Example: Columns: length, first_word, last_word Prompt: Count the characters and return as 'length'. Extract the first word as 'first_word' and the last word as 'last_word'.

Custom (LLM Modifier)

A flexible LLM-powered modification service that transforms column values based on a custom prompt. Use the LLM to intelligently clean, normalize, rewrite, anonymize, extract or restructure data.

LLM model: gpt-4o-mini

Input: Column values to modify, the operation to perform and custom instructions describing how to modify them.
Output: Updated column values, a new joined column, or multiple new columns generated by a split.

How to use:

  • Choose an operation type (Join / Split / Edit directly).
  • (Optional) Select additional columns for the join operation.
  • (Optional) Rename new generated columns for the join and split operation.
  • Write instructions describing how to modify the values. The LLM will receive each cell value and return output based on the operation type.

Example propmts:

  • Join: Merge first_name and last_name into a full name. Trim spaces and capitalize properly.
  • Split: From 'address' extract street, city, and postal_code. Return values exactly as: street | city | postal_code.
  • Edit directly: Standardize dates to ISO format (YYYY-MM-DD). If missing, return an empty string.