TalentMatch AI: LLM-Powered Resume Screening

TalentMatch AI overview dashboard displaying the system workflow for automated resume screening and talent matching.
Candidate leaderboard ranking five sample candidates based on match scores, matching skills, and missing skills for a selected job role.
Recruitment session configuration interface for selecting a predefined job role, entering job requirements, and choosing a screening mode.
Resume upload interface supporting multiple PDF resumes and sample candidate profiles for demonstration and screening.
Candidate screening results displaying ranked candidates, match scores, skill matches, skill gaps, and export options.
Model evaluation dashboard showing performance metrics and ranking accuracy.
System architecture of TalentMatch AI — LLM-Powered Resume Screening and Talent Matching Platform.

Inspiration

Campus placement teams often receive numerous resumes for different job roles. Manually reviewing every resume and comparing candidate skills with job requirements can be time-consuming. Traditional keyword-based screening may also overlook relevant skills when candidates describe their experience using different terms. This inspired us to develop TalentMatch AI, an AI-powered platform that helps streamline resume screening and candidate evaluation.

What it does

TalentMatch AI is an LLM-powered resume screening and talent matching platform designed for campus placements. Users can select a predefined job role or enter a job description, upload multiple candidate resumes in PDF format, and evaluate candidates against the selected requirements.

The platform extracts relevant candidate information, compares skills with job requirements, calculates match scores, identifies matching and missing skills, and ranks candidates on a leaderboard. It also provides candidate details, supports exporting rankings to Excel and CSV, and includes a model evaluation dashboard for assessing performance using a manually labelled benchmark dataset.

How we built it

We developed the platform as a web application with a dashboard-based interface and a step-by-step recruitment workflow. The system supports recruitment session setup, multiple resume uploads, candidate screening, ranking, and model evaluation.

The screening workflow uses text processing, TF-IDF-based representations, cosine similarity, skill matching, and qualification-related information to evaluate candidate profiles against job descriptions. The results are presented through match scores, skill-gap indicators, and a ranked candidate leaderboard. A separate evaluation module reports classification and ranking metrics using manually labelled resume–job description pairs.

Challenges we ran into

One of the main challenges was comparing resumes with job descriptions when candidates used different terms to describe similar skills. Another challenge was combining semantic similarity and skill coverage into a meaningful candidate score while clearly showing why a candidate matched or missed specific requirements.

We also needed to maintain candidate rankings across screening sessions and evaluate the system using labelled examples. Ensuring that the evaluation results were interpreted correctly was important because the available benchmark dataset was limited.

Accomplishments that we're proud of

We successfully developed a working prototype with job-description selection, multiple PDF resume processing, candidate scoring, skill-gap analysis, ranked candidate results, and export functionality.

We also implemented a model evaluation dashboard. On our manually labelled benchmark of eight resume–job description pairs across two campus job positions, the system reported 100% precision, 40% recall, and an F1-score of 57.1%. Top-1 and Top-3 ranking accuracy were both reported as 100% on the tested benchmark cases. These results provide an initial evaluation of the prototype rather than a guarantee of performance on larger datasets.

What we learned

Through this project, we learned how text processing, semantic similarity, skill matching, and ranking can be combined to support resume screening. We also gained experience in designing a complete application workflow, presenting explainable results, and evaluating a prototype using precision, recall, F1-score, and ranking metrics.

An important lesson was that high precision alone does not guarantee that all relevant candidates will be identified. The recall result highlighted the need to improve candidate coverage and validate the system using a larger and more diverse dataset.

What's next for TalentMatch AI: LLM-Powered Resume Screening

Our next steps include expanding the evaluation dataset, testing the platform across more job roles, improving the identification of relevant candidates, and refining the matching and ranking methods. We also aim to improve the explanations provided for candidate scores and strengthen evaluation of semantic matching and skill-gap detection.

In the long term, TalentMatch AI could be extended with additional recruitment workflows and human-review support, helping placement teams make more informed and transparent initial screening decisions. The platform is intended to assist human decision-makers rather than replace them.