---

Introduction to the Problem

On one of our workdays, while discussing the results of testing a new feature with the team, we noticed that users often encountered issues when searching the database. This led to frustration and increased query execution time. One of the developers raised the question: "Why not try using GPT for data parsing instead of searching?"

Context of the Problem

The issue with database search was relevant for the entire team. Users looking for information about job vacancies or candidates complained about the slowness and inaccuracy of the results. This was critical, as the quality of the search directly affected how successfully we could help companies find the right candidates on /jobs and /for-companies. We understood that solving this problem could significantly enhance user experience and increase customer satisfaction.

Details of the Problem

Specifically, we faced situations where user queries often returned irrelevant results. For example, one user searched for a candidate with Java experience but received a list dominated by results related to Python. This not only frustrated users but also diminished trust in our platform.

Initial Attempts to Solve the Problem

Initially, we attempted to improve the search algorithm by using more sophisticated filtering and ranking methods. However, after several iterations and tests, we realized that this did not address the core issue: the efficiency of database search depended on the limitations of the data structure itself. We also considered using advanced indexing, but this would require significant infrastructure costs and time for implementation.

Technical Approach to the Solution

Ultimately, we concluded that moving GPT functionality to the data parsing stage would be more effective. We began integrating GPT to analyze user queries and extract the necessary information from the database.

import openai

response = openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[{"role": "user", "content": "Find a candidate with Java experience"}]
)

This approach allowed us to process queries more flexibly and accurately, improving response quality and reducing response time.

Changes in the Product

After integrating GPT into parsing, we noticed significant improvements in the quality of results. Users began receiving more relevant outcomes, positively impacting user experience. We also started receiving positive feedback from companies using /for-companies, noting that finding suitable candidates had become easier and faster.

Lessons Learned

• Understanding data structure is critical for choosing the right search approach.
• Using GPT for data parsing can significantly improve result accuracy.
• One should not hesitate to abandon initial solutions if they are not yielding results.
• User feedback should be a primary criterion for evaluating changes.

What This Means for Candidates

For candidates, this means they can now expect more accurate results when searching for job vacancies on our platform. The improved parsing system helps them find suitable offers more quickly, enhancing their chances of successful employment.

What This Means for Recruiters

For recruiters, this provides the opportunity to find the right candidates more efficiently. With the improved parsing system, they can quickly identify suitable specialists, reducing the time to fill vacancies and increasing overall productivity.

Next Steps

However, we understand that the work is not finished. We continue to monitor how users interact with the new system and gather data on how it affects their experience. If we had to go through this process again, we would test various integration approaches more thoroughly before making a final decision. In the future, we plan to explore opportunities for improving machine learning algorithms for further optimization. ---