Finding a valuable job is a life goal for most of the people in the world. But not only employees have the challenge to work at the best possible company. Also, employers are facing the problem of matching the best candidate for the role they offer. Our client, Work Service decided to support companies which are struggling with such an issue. Its goal was to create the best matching system.





Work Service is the biggest HR company in Poland. It specializes in the field of personnel consulting, as well as restructuring in the area of HR, recruitment and employee outsourcing. It wanted to stand out by providing innovative solutions for employers and employees.




Work Service wanted to speed up the recruitment process by preparing a recommendation platform connected with resumes. Job offers are sorted and matched with a potential candidate. It helps to automate recruiters’ work and save time which can be dedicated to different activities in the company. One of the challenges was to exclude candidates without needed experience, educational background or skills.





Based on this need, Stermedia created a system processing job offers in batch mode.  We analyzed all of the possibilities and focused on:


  • geolocalization
  • preparing map/graph of the skills, education, and roles
  • creating the application to labeling from CVs and job offers
  • automatic data extractions (like skills, education, and experience) from CV
  • candidates recommendation system working as Machine Learning in both ways:
    job offer -> candidate and candidate -> job offer


  • initial phone job interview with the candidate run by an AI bot; based on it, extra information about the candidate is collected effortlessly
  • AI bot is connected with RMS system which has been developed in Java by Stermedia since 2015

As the result can be distinguished CV/Job offer skills extraction in followed numbers:

  • 6000 skills
  • 3000 positions
  • 3000 education entities
  • 10 models of artificial intelligence
  • 100 000+ models fitted for parameters optimization




Recommender system – technically

    • Text lemmatization and normalization
    • HTML tags treatment
    • Misspelling treatment
    • Word2Vec (a numerical representation of text)
    • Brand classification of candidate
    • Skills extraction


Python, k-means, knn, NEF, LDA, PCA, Factorization Machines, xgBoost, word2vec, levenshtein distance, django, Django Rest Framework, d3.js, ggplot

Recommender system in models:

  • Clustering (k-means)
  • Topic modeling (NEF, LDA)
  • Classification (xgBOOST)
  • Recommendation (FM, xgBOOST)




‘Premium team is undoubtedly the main asset of Stermedia. Their developers have a problem-solving and proactive attitude. Unlikely to some other companies, communication with them was smooth and efficient. It significantly speeded up the end of the project.” – said Piotr Adamczyk, CIO, Work Service S.A.



Time Hours 1512
Months 6
Team size Total 5
Maximum 4
Approximately 3