When we set out to build Placr's AI matching engine, we faced a fundamental tension. The most accurate machine learning models are often the least interpretable. Black-box models can tell you that Candidate A is a 92% match for a role, but they cannot tell you why. For recruitment agencies whose entire value proposition is expert judgment, a score without explanation is worse than no score at all.
We chose a hybrid architecture that combines semantic embeddings for deep understanding with structured feature extraction for explainability. Every match score in Placr is accompanied by a detailed breakdown: which skills matched, what experience is relevant, where the gaps are, and how the candidate compares to the ideal profile. Recruiters don't just see a number. They see the reasoning.
The technical foundation is a multi-stage pipeline. First, we parse CVs and job descriptions into structured representations using Anthropic's Claude, with every response validated against a strict schema. Nothing is fine-tuned; the structure comes from the schema, not from a custom model. Then a deterministic scoring engine compares the candidate with the role across multiple dimensions, including semantic similarity, skills, experience, location, and salary. Each dimension gets its own score and a plain-language explanation, so you can see which ones carried the result.
The point of all this is trust. A recruiter who can see why a candidate scored the way they did can agree, disagree, or overrule the system, and a score they can interrogate is one they will actually use. Trust drives adoption, adoption drives efficiency, and efficiency drives placements. That is the virtuous cycle we designed for.
