Skip to main content
When a mapping job finishes, some rows always come back unmatched. This recipe configures an agent to search the golden dataset for each one and file the top N candidates as pending items for a human reviewer.

What you’ll build

An agent that, on every MappingJobFinishedEvent:
  1. Pulls the job’s unmatched results.
  2. Calls search_golden_dataset (by name) and search_golden_dataset_by_geo (by coordinates) per row.
  3. Files a MATCH_PROPOSAL pending item with the top 3 candidates.
A human then opens Pending, picks a candidate (or rejects all), and the link is made.

Prerequisites

  • Pro Max workspace.
  • At least one historical mapping job with unmatched rows.

Steps

1

Create the agent

Agents → New Agent. Pick the starter template “Triage unmatched rows”. It pre-fills the goal and rules.
2

Goal

Watch every mapping job. For each unmatched row, search the golden dataset by name and by geo, and propose the top 3 candidates as pending items.
3

Scope

Optional — narrow to a specific supplier (e.g. BOOKING) if you only want this for one source.
4

Rules

5

Activate

Click Activate. The next mapping job that finishes will trigger the agent.

Equivalent API call

Then POST /agents/{id}/rule-sets with the JSON above and POST /agents/{id}/activate.

Reviewing pending items

Pending tab with match proposals Each MATCH_PROPOSAL item shows:
  • The unmatched partner row (name + address).
  • 3 candidate reference hotels with referenceHotelId, name, distance, and the agent’s 1-sentence reasoning.
  • Accept (links the result) / Reject (dismisses) buttons.