What is semantic search and why does it matter?

Semantic search uses 384-dimension vector embeddings to understand meaning, not just keywords. If you search for 'backend engineer,' it also finds candidates who describe themselves as 'server-side developer' or 'API architect', because the meaning is similar even if the words are different. This dramatically improves recall over traditional keyword-only search.

What is Reciprocal Rank Fusion (RRF)?

RRF is a ranking algorithm that combines results from multiple search methods. Instead of relying on one approach, it merges ranked lists from text search, semantic search, entity search, and LLM-based search. Candidates who rank consistently high across all four methods get the best final position, producing more accurate results than any single method alone.

What entity types are extracted from resumes?

Five named entity types are extracted and indexed: SKILL (programming languages, tools, frameworks), COMPANY (employer names), ROLE (job titles held), INSTITUTION (universities, certifications), and PROJECT (specific products or initiatives the candidate worked on). These entities power structured search filters.

What happens if resume parsing fails?

The processing pipeline automatically retries failed resume parsing. If a resume cannot be parsed after retries, it is flagged for manual review. The candidate profile is still created, you just will not have AI-extracted structured data until the issue is resolved.

Is AI search included in the free plan?

Yes. Resume parsing, suitability scoring (0-100), 384-dimension vector embeddings, named entity extraction, and hybrid search with Reciprocal Rank Fusion are all included on every plan, including the free Starter plan with 5 active jobs and top 50 candidates per job.

AI-Powered Search That Understands Context

From Raw Resume to Structured Intelligence

Every resume uploaded to Curriculo ATS goes through a multi-stage AI pipeline that extracts, scores, and indexes candidate data.

Resume Parsing — AI extracts structured data from PDF resumes: candidate name, email, work experience (company, role, dates, descriptions), education (institution, degree, dates), and skills. No manual data entry required.
Suitability Scoring — Each candidate receives an AI-generated suitability score from 0 to 100%, along with written reasoning explaining why the score was assigned. The score evaluates fit against the job requirements, not just keyword overlap.
AI Summary & Auto-Tags — A concise AI-generated summary highlights the candidate's strengths and key qualifications. Auto-tags are applied based on extracted skills, experience level, and industry, making filtering instant.
Pipeline Recovery — If any stage of AI processing fails (network timeout, model error), the pipeline auto-retries. Failed candidates are queued for reprocessing so nothing gets stuck in a broken state.

Vectors, Entities, and Semantic Understanding

Vector Embeddings — Every resume is converted into a 384-dimension vector embedding using pgvector. This captures the semantic meaning of the entire document, so searching for "backend engineer" also finds candidates who describe themselves as "server-side developer" or "API architect."
Segment Embeddings — Beyond the full-resume embedding, separate vectors are generated for each section: experience, education, and skills. This lets you search within specific resume sections for more precise results.
Entity Extraction — Named entities are extracted and classified into five types: SKILL (Python, React, SQL), COMPANY (Google, Stripe), ROLE (Senior Engineer, Product Manager), INSTITUTION (MIT, Stanford), and PROJECT (specific projects or products). These entities power structured search filters.

Four Search Methods, One Ranked Result

Curriculo ATS runs four search methods in parallel and merges the results using Reciprocal Rank Fusion (RRF) for the best possible ranking.

Text — Full-text search across resume content and parsed fields
Semantic — Vector similarity search via pgvector embeddings
Entity — Structured search across extracted SKILL, COMPANY, ROLE entities
LLM — AI re-ranking using LLM judgment for nuanced relevance

Reciprocal Rank Fusion & Gmail-Style Filters

Reciprocal Rank Fusion (RRF) — Each search method produces its own ranked list. RRF combines these lists by assigning scores based on rank position across all methods. Candidates who appear high in multiple lists get the strongest final score, even if no single method ranked them #1. This produces more reliable results than any single search approach alone.
Gmail-Style Scope Filters — Narrow your search with familiar scope operators: in:rejected to search rejected candidates, in:trash for archived applicants, in:shortlisted for your shortlist. Combine with any search query for fast, precise filtering across your entire candidate pool.

Stats

384 — Dimensions per vector embedding
5 — Entity types extracted per resume
4-way — Hybrid search with RRF fusion
$0 — AI search included on every plan