Suggest peer reviewers for new manuscripts in Airtable

Build me an agent workflow that suggests peer reviewers whenever a new manuscript is added to my Airtable 'Manuscripts to Review' table.

Trigger: Airtable poll on the 'Manuscripts to Review' table with event type new_record. When the trigger fires, read the new row's title, abstract, submitting authors (list of names), and submitting institutions (list of names) from Airtable.

Step 1 — find topically related work in OpenAlex. Call OpenAlex Search Works with a semantic search on the manuscript's abstract (search.semantic=<abstract>), filtered to the last 5 years (from_publication_date). Ask for around 40 results, sorted by relevance. If the abstract is empty or the semantic call returns fewer than 20 works, fall back to a title.search on the manuscript title with the same 5-year window. Select fields id, display_name, doi, publication_year, cited_by_count, and authorships so we get author IDs, names, ORCIDs, and institution IDs inline.

Step 2 — build the candidate pool. Collect the authorships across those related works. Dedupe by OpenAlex author ID. For each candidate, remember (a) how many of the related works they appeared in (topical-fit count), and (b) the highest-cited related work they authored (title, doi, cited_by_count) for use in the rationale later.

Step 3 — enrich each candidate. For each unique author, call OpenAlex Get Author to pull works_count, cited_by_count, h_index (or 2year_mean_citedness as the seniority/impact proxy), ORCID, and their most recent institution. Then call OpenAlex Get Institution on that institution ID to resolve the display_name and country. Skip any candidate that has no ORCID or no current institution — editors need contactable, verifiable people.

Step 4 — conflict-of-interest filter. Drop a candidate if any of the following are true: (a) their OpenAlex author ID or name matches one of the submitting authors, (b) their current institution's display_name matches (case-insensitive, ignoring punctuation) any of the submitting institutions, or (c) they have co-authored a work with any submitting author (check by looking up submitting authors in OpenAlex and comparing coauthor sets, or by scanning the candidate's authorships for a shared work with a submitting author).

Step 5 — rank. Score the remaining candidates using a blend of: topical-fit count (how many of the related works they appeared in), recent citation impact (cited_by_count or 2year_mean_citedness), and seniority (works_count / h-index proxy). Sort descending. Keep the top 8 to 10.

Step 6 — write the shortlist back to Airtable. For each of the top candidates, use Airtable Create Records to append a row to the 'Suggested Reviewers' table in the same base with fields: name, affiliation (institution display_name), ORCID, cited_by_count, most_cited_related_paper (title + DOI link), and rationale (one plain-English sentence explaining why they fit, referencing their most-cited related paper and topical overlap). Also include a manuscript link field (multipleRecordLinks back to the source row) if that field exists on the table. Batch the create calls in groups of 10.

If no candidates survive the filters, write a single row into 'Suggested Reviewers' with name = 'No candidates found' and rationale explaining what was tried (semantic search hit count, how many were filtered by COI, how many lacked ORCID/institution) so the editor knows to widen the criteria.

Keep the rationale sentences short and specific — mention the topical overlap and the most-cited related paper. Don't fabricate emails; ORCID + affiliation is enough for the editor to reach out.