Explainable ATS Matching Engine
Elevra
An ATS match-analysis system that combines deterministic keyword matching, semantic retrieval, evidence classification, and AI-generated explanations to evaluate how well a resume satisfies a job description.
Engineering approach
- Keyword matching
- Semantic embeddings
- Requirement-level retrieval
- Evidence classification
- Contextual reranking
- Corroboration
- Calibration
- Deterministic score aggregation
- Gemini explanation layer
- Python
- NLP
- Sentence Transformers
- Semantic Retrieval
- Gemini
- PDF Parsing
- FastAPI
System architecture — tap a stage for detail
- Parse & retrieve evidencekeyword + semantic
Candidate uploads a resume as a PDF.
Raw text is pulled from the PDF for downstream parsing.
Skills, education, and experience are structured into discrete fields.
The job description is parsed into required and preferred requirements.
Presence check — does each required skill literally appear in the resume?
Requirement-level embedding search finds the resume's best-supporting evidence for each requirement.
Each match is classified — demonstrated, skill-list mention, aspirational, or irrelevant — not just scored by similarity.
- Deterministic enginethe system calculates
Keyword and evidence-weighted scores combine into one reproducible, auditable number. No LLM involved.
- Explanation layerthe LLM explains
Reads the already-computed score and writes the human-readable explanation. It never changes the result.
- Output
Displays the score, evidence, and explanation to the candidate or recruiter.
“Gemini explains the decision.It does not make the decision.”