From Incomplete Candidate Profiles to More Relevant Job Matches

A candidate can have the right experience for a role and still be difficult to find.

In many recruitment platforms, candidate discovery depends heavily on structured profile fields, exact keywords and manually defined search rules. When someone uploads a detailed CV but leaves parts of their profile incomplete, valuable information about their experience may never influence the search results.

For one recruitment marketplace, this created a gap between the information available about candidates and the information its matching system could actually use.

We implemented an AI-powered candidate matching service that combines structured recruitment criteria, information extracted from PDF CVs and semantic analysis of candidate experience.

The proof-of-concept engine evaluates candidate relevance against job requirements, considers information beyond exact keyword matches and produces ranked recommendations for recruiter review.

Rather than requiring every relevant qualification to appear in a perfectly completed profile, the system provides a more context-aware way to discover candidates already represented in the platform's data.

The recruitment marketplace allowed candidates to create profiles, upload CVs and apply for jobs.

Employers and recruiters could search the available talent pool using structured fields and predefined criteria.

But the information candidates entered into profile forms did not always represent their complete professional experience.

Candidate information was split between profiles and CVs. A structured profile might contain a job title, location and selected skills, while the uploaded CV described detailed responsibilities, previous roles and relevant professional experience. When matching relied primarily on structured fields, much of that additional information could remain unused. A candidate could be relevant to a vacancy without appearing in the expected search results.

Keyword matching struggled with differences in language. Recruiters and candidates do not always describe the same expertise using identical terminology. A job description might request one combination of skills and responsibilities, while an experienced candidate describes similar work using different titles or industry-specific language. A rigid keyword-based system can struggle to recognize that relationship. The challenge was to evaluate the meaning of candidate experience, not simply the presence of particular words.

Fixed matching rules couldn't capture every recruitment requirement. Recruitment involves both objective eligibility criteria and contextual professional judgments. Location, remote-work preferences and other predefined requirements may need to be applied consistently. At the same time, seniority, relevant experience and domain expertise often require interpretation. The existing approach did not bring those different kinds of evaluation together effectively.

The matching engine needed to evolve without disrupting the existing platform. The business already had a recruitment marketplace and established candidate records. It did not need to replace the platform or rebuild its entire recruitment process. The objective was to develop a matching capability that could work with existing candidate information, interpret CV content and support more relevant recommendations through a modular integration approach.

We implemented a standalone proof-of-concept matching service designed to evaluate candidates against job requirements using multiple sources of information.

The solution separates document interpretation, eligibility filtering and contextual scoring so that each contributes to the matching process.

AI-powered CV information extraction: The engine processes supported PDF CVs to identify information relevant to candidate evaluation. This includes professional experience, seniority indicators, skills and domain knowledge described in the document. The extracted information supplements the candidate's structured profile instead of relying only on fields manually completed during registration. This gives the matching process access to relevant qualifications that may otherwise remain hidden inside an uploaded CV.

Configurable recruitment filtering: We implemented a structured filtering component that applies defined recruitment criteria to candidate information. The rules can consider factors such as location, remote-work eligibility and other requirements relevant to the role. Where supported, the filtering process uses both existing profile information and information extracted from the CV. This allows explicit eligibility conditions to be considered alongside broader contextual matching.

Context-aware candidate evaluation: A language-model component interprets the professional meaning of candidate experience in relation to the job description. Instead of depending entirely on exact words, the system considers how responsibilities, qualifications, seniority and professional background relate to the requirements of the role. This supports more nuanced evaluation where similar experience may be described using different terminology.

Vector-based relevance scoring: The matching engine uses semantic representations of candidate information and job requirements to calculate relevance scores. These scores help organize candidate recommendations according to their relationship with the vacancy. Structured filtering and contextual scoring work together, rather than treating semantic similarity as the only factor in candidate suitability. The resulting ranking is decision support for recruiters, not a final hiring judgment.

Standalone architecture designed for platform integration: The matching engine was implemented as a modular service with an API-oriented design. This allows the matching capability to be developed and tested separately from the recruitment marketplace's main application. The proof of concept was evaluated using a separate non-production copy of the relevant data environment, avoiding the need to modify live recruitment workflows during development. The architecture provides a foundation for future integration and expansion to additional job categories, subject to further validation and deployment decisions.