ºÚÁÏÃÅ

Bolt.Works is a Finnish technology company specializing in staff recruitment. The Bolt.Works app has created a new marketplace for the workforce and its subscribers, connecting employees with jobs in the construction, manufacturing, logistics, property management, hospitality and retail sectors. The million dollar question for Bolt.Works’ business was how to enable scalable growth internationally.

Ìý

Ìý

Laying foundations to scale and expand

Ìý

An experienced recruiter can skim unstructured texts like CVs and job ads to match potential jobs with workers. However, this is a manual process, and not highly scalable. In order for Bolt.Works’ business to expand and grow at scale, locally and internationally, matching automation was needed.

Ìý

Automating the matching process would enable recruiters to focus on higher-value activities, such as prioritizing which workers to progress, and eventually select, for the role. These higher-value decision making tasks, which benefit from human knowledge and experience, should not be automated, but rather augmented with AI.

Ìý

Ìý

Matching jobs with workersÌý

Ìý

Bringing years of experience in Data and AI solutions, ºÚÁÏÃÅ partnered with Bolt.Works to co-deliver an AI-based matching engine for jobs and workers to enhance, scale and speed up the capabilities of their platform and business.

Ìý

A matching engine powered by AI allows Bolt.Works to faster fulfill their clients’ worker needs. It also benefits the potential employee by automatically suggesting relevant jobs for them to apply for. Moreover, multilingual AI enables Bolt.Works to serve clients internationally.

Ìý

The AI powered matching engine also enables Bolt.Works to gain new data insights, like what kind of worker groups they have, about the job-to-worker matching process, which is valuable for further enhancing customer experience and processes.

Ìý

Ìý

Transforming unstructured data into structured with generative AIÌý

Ìý

Natural Language Processing (NLP) is a branch of AI that focuses on enabling computers to understand, interpret, and generate human language, like job ad texts, by developing algorithms and models. NLP methods have evolved from classical exact term-based matching solutions, to the latest deep learning-based large language models (LLMs).

Ìý

ChatGPT is a generative large language model developed by OpenAI that can generate human-like responses and engage in conversational interactions. In this project, we successfully leveraged the Microsoft Azure OpenAI Service ChatGPT API to extract and generate structured data from unstructured texts of jobs and job applications.

Ìý

Utilizing ChatGPT saved weeks of development time compared to building traditional NLP information extraction models. It also allowed us to generate new data, like summaries of longer texts which enabled building a better working matching engine solution.

Ìý

Ìý

Semantic similarity matching with large language models

Ìý

ChatGPT alone is not enough to match jobs and workers. So we fine-tuned an open-source, multilingual large language model which learned to match semantically similar jobs and workers based on the structured data generated by ChatGPT.

Ìý

We also laid the groundwork for future Reinforcement Learning - so the matching engine can continuously explore and learn from the recruiter’s and workers’ feedback. We call this approach a Decision Factory.

Ìý

Ìý

Responsible AI

Ìý

Generative AI, like ChatGPT, has issues, like the potential to generate biased, discriminatory or false content. In recruiting, ethical concerns are very prominent and regulated. We addressed this by only using ChatGPT to extract non-sensitive structured data, and carefully guiding and validating its outputs. Final matching is not done by ChatGPT, but by a model that is completely under our control. This enables us to mitigate biases and, for example, exclude information related to gender and ethnicity.

Ìý

We discussed the next steps to ensure that the outputs of the matching engine are explainable for human recruiters, to help them make their final recruitment decision.

Ìý

To ensure GDPR compliance, we implemented the Microsoft Azure version of ChatGPT which is hosted on Azure servers physically located in the EU and isolated from OpenAI.

Ìý

Ìý