---

The Beginning: A Conversation in Slack

On an ordinary day in our Slack, we discussed how to improve the recruitment process for companies. One of the developers mentioned that candidates often struggle to understand employers' requirements. This moment became a starting point for our team.

Context of the Problem

We worked in the HR technology sector and understood how crucial quality recruitment is for business growth. Companies complained about the lack of suitable candidates, while job seekers faced a lack of transparency in requirements and expectations. Every day we saw how this gap created problems for both sides. It was essential to address this issue to improve processes for both candidates and employers.

The Specific Problem

One day we received a request from a startup that could not find a suitable developer for their team. They posted a job opening but received numerous resumes from candidates who did not meet the requirements. This led to wasted time for both HR and applicants. We realized that the current tools were unable to effectively match candidates' skills with companies' needs.

Initial Attempts

Our initial attempts involved using standard resume matching algorithms. We thought that simply increasing the amount of data would improve the results. However, this did not yield significant effects. We encountered issues with data redundancy and lack of context, making the algorithm ineffective. Eventually, we decided to rethink our approach and focus on a deeper understanding of skills and requirements.

Technical Approach

Our team developed a new algorithm that considered not only keywords but also contextual relationships between skills and requirements. We implemented semantic matching, which allowed for more accurate candidate assessments. Here is a snippet of code that we used to analyze resumes:

import spacy

def analyze_resume(resume_text):
    nlp = spacy.load('en_core_web_sm')
    doc = nlp(resume_text)
    skills = [token.text for token in doc.ents if token.label_ == 'SKILL']
    return skills

This approach enabled us to more accurately identify skills and matches, ultimately reducing the time spent on recruitment.

Changes in the Product

After implementing the new algorithm, we noticed a significant improvement in the quality of recruitment. Companies began receiving more relevant resumes, while candidates received offers that matched their skills. We updated the pages on our website to reflect these changes and enhanced the /jobs and /for-companies sections, emphasizing the accuracy of recruitment.

Lessons Learned

• Understanding context is more important than simply increasing data volume.
• Utilizing semantics can significantly enhance matching accuracy.
• The team must be open to mistakes and failures — this is an important part of the process.
• Effective communication between HR and developers is critical for success.
• The importance of user feedback should not be underestimated.

What This Means for Candidates

For candidates, this means they can expect a more personalized and accurate approach to job matching. We aim to provide clearer information about employers' requirements and expectations, which in turn improves their chances of successful employment.

What This Means for Recruiters

Recruiters can now work with higher quality data about candidates, which means less time spent filtering out unsuitable resumes. This will allow them to focus on more strategic aspects of recruitment, such as engaging with candidates and building relationships with clients.

Next Steps

We continue to monitor the results of our algorithm and plan further improvements. In particular, we want to explore integration possibilities with other platforms and data. If we had to change anything, we would start with a deeper analysis of user needs at the very early stage. ---