Candidate Search Retrieval: A 7-Point Release Checklist for Fresh, Citable Results

The article argues that candidate search should be treated first as a retrieval product rather than an LLM feature, with the answer a recruiter needs defined before any system design begins. It proposes a retrieval contract covering candidate identity, matching evidence, tenant scope, recency, and source identifiers. A vector index serves as durable memory for owned content, while live web search covers recent public material, and the two should not silently substitute for each other. The author recommends a two-lane pipeline that filters by tenant and access metadata before ranking, merges results by candidate ID, and labels which evidence came from durable versus live sources. An illustrative TypeScript example shows bounded requests with rate-limit retries, and the article notes the right tooling choice depends on ownership and operating constraints.