Job Matching

Matching should be explainable and grounded in structured data. The system should not rely on embeddings or model guesses alone.

Matching goals

Compute a job score from a combination of:

• semantic similarity
• skill overlap
• experience compatibility
• seniority compatibility
• location compatibility
• work mode compatibility
• salary compatibility
• candidate preferences
• domain relevance

Rules

• Start with deterministic scoring from structured CandidateProfile and JobListing objects.
• Add LLM reasoning only where it improves explanation or handles ambiguity.
• Every score must explain why it was high or low.
• Keep weights configurable and measurable.
• Never claim a candidate has a skill or qualification unless it is supported by validated profile data.

Output format

Each match should return a score and an explanation such as:

Job Match: 87/100
Skills: 92% | Experience: 88% | Seniority: 95% | Semantic: 84%
Strong: Python, FastAPI, SQL, REST APIs
Missing: Kubernetes, AWS

Repo principle

Deterministic logic should handle the bulk of matching. The LLM is an assistive layer for explainability and edge cases, not a source of fabricated experience or invented qualifications.