Skip to main content

ResDB Talent Pool & Semantic Sourcing

ResDB is the neural candidate database powering XeCubes. Unlike legacy ATS databases that rely on rigid keyword matches, ResDB indexes every resume, past interview transcript, and project repository into high-dimensional vector space using fine-tuned embedding models.


How Semantic Search Works​

When a recruiter enters a natural language query or imports a job description, ResDB computes cosine similarity between the query embedding and candidate vectors:

flowchart TD
JobDesc[Job Description or Natural Language Query] --> Embedder[XeCubes Neural Embedding Engine]
Embedder --> Vector[1536-Dimensional Query Vector]
Vector --> PgVector[(PostgreSQL + PgVector HNSW Index)]
PgVector --> TopK[Top-K Nearest Neighbor Candidate Profiles]
TopK --> ReRanker[LLM Re-Ranking & Calibration]
ReRanker --> RankedMatches[Ranked Candidates with Explanation Badges]
  • Synonym & Domain Mastery: Recognizes that "Distributed systems with Paxos/Raft" implies strong Go/Rust or C++ background even if specific words are missing.
  • Talent Rediscovery: Unlocks dormant candidates from past recruitment campaigns who previously reached final rounds.
  • Zero Keyword Stuffing Sensitivity: Context-aware embeddings evaluate depth of contribution rather than keyword repetition count.

Sourcing Filters & Boolean Overlays​

ResDB allows hybrid querying combining semantic vectors with structured metadata:

{
"semantic_query": "Senior Site Reliability Engineer with high-traffic Kubernetes and eBPF tracing experience",
"filters": {
"locations": ["San Francisco, CA", "London, UK", "Remote"],
"min_experience_years": 5,
"max_notice_period_days": 30,
"skills_required": ["Kubernetes", "Linux Kernel", "Go"],
"salary_expectation_max": 220000
},
"top_k": 25
}

Candidate Explainability​

Every candidate returned by ResDB includes an AI Match Scorecard:

  1. Core Competencies Match: Direct alignment with required technical stack.
  2. Growth Trajectory: Progression of seniority and scope of ownership in past companies.
  3. Identified Gaps: Missing certifications or technologies requiring ramp-up time.
  4. Suggested Interview Questions: Dynamic questions generated specifically to validate candidate strengths or probe gaps.