krzysztof/landing_page
1
0
Fork 0
mirror of https://github.com/qdrant/landing_page.git synced 2026-10-04 02:18:29 +02:00
Code Issues Packages Projects Releases Wiki Activity

Update case-study-go-perfect.md

Browse Source
This commit is contained in:
daniel-azoulai
2026-05-18 13:29:19 -07:00
parent 29442b0bd2
commit 99c2fec65d
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
qdrant-landing/content/blog/case-study-go-perfect.md
+1 -1
View File
@@ -71,7 +71,7 @@ Within GoPerfect's user-facing interactive, sub-agent-loop latency budget, the a
### How it works in production
The production pipeline runs in three layers. Ingestion enriches profile data from multiple sources (professional networks, code repositories, company data, and AI-derived signals) and writes structured multivector points into Qdrant. Retrieval combines hybrid search with category-specific multivector queries to return a high-confidence candidate pool. An LLM orchestration layer above Qdrant runs the agent loop: it generates sub-questions, issues parallel searches, evaluates the returned candidates, and assembles the final ranked shortlist.
The production pipeline runs in three layers. Ingestion enriches profile data from multiple sources (professional networks, code repositories, company data, and AI-derived signals) and writes structured multivector points into Qdrant. Retrieval combines hybrid search with multivector queries to return a high-confidence candidate pool. An LLM orchestration layer above Qdrant runs the agent loop: it generates sub-questions, issues parallel searches, evaluates the returned candidates, and assembles the final ranked shortlist.
GoPerfect also runs a candidate scoring and fraud-signal layer that cross-references claims in a resume against external evidence. A candidate who lists a programming language, for example, gets validated against their public code activity. The scoring layer also learns from outcomes: when a candidate flagged with a fraud signal goes on to get hired through normal channels, the system weights similar signals more leniently going forward. Recruiters see a graphical view of each candidate that includes both the ranking and the supporting evidence, with adjustable tolerance for fraud signals depending on the role context.