Powering Career Coaches with Large Language Models 🤖
Finding the right job can be difficult, especially for individuals who lack the necessary expertise and experience.
On the other hand, career coaches often face obstacles when assisting their clients during the job search process.
To overcome these challenges, we collaborated with Workforce Singapore (WSG) to develop the Job Adjacency Tool (JAT).
By leveraging Large Language Models (LLMs), JAT equips career coaches with the insights they require to offer personalised guidance to their clients.
Superheroes Of The Job Market: Career Coaches
Career coaches are professionals who offer guidance and support to individuals who are seeking to navigate their career path or make mid-career transitions.
But even superheroes have their kryptonite.
The process of scouring various sources, including job portals like MyCareersFuture and their clients’ resumes, to find the perfect job match can be arduous.
It becomes even more challenging when career coaches need to locate jobs that require specific skillsets or are in a particular industry.
In addition, career coaches need to research companies and industries to ensure that they align with their clients’ goals and values.
Although our coaches are experts in some industries, they may not have domain knowledge of all industries and sectors.
This makes it challenging to provide tailored career guidance to their clients seeking jobs in unfamiliar professions.
Innovation to Automation
Manual processes in coaching tasks often lead to mistakes and inconsistencies that can negatively affect the relationship between the career coach and their clients.
To address this, we aim to automate certain tasks that are time-consuming and labor-intensive for the coaches.
Our primary objective is to allow them to focus more on understanding the unique skills, interests, and career goals of their clients
By doing so, we hope to achieve better results for both jobseekers and coaches, leading to more successful job placements.
JAT Version 1: A Sidekick for Career Coaches
JAT Version 1, powered by the OpenAI GPT-3 engine, was developed to recommend relevant and adjacent job roles for jobseekers.
The aim was to suggest job roles for jobseekers based on their skills and experience.
Although the accuracy of the application was questionable, the team saw its potential and enlisted career coaches for validation.
JAT Version 2: Not Quite JARVIS Yet
Our initial version of JAT received positive feedback.
And with JAT Version 2, we introduced new features to improve our application.
One enhancement that we implemented was to provide detailed job recommendations and better descriptions of why a jobseeker was a good fit for a particular role.
To facilitate this feature, we upgraded the model from GPT-3 to GPT-3.5.
GPT-3.5 offered an increased token limit from 2048 to 4096 as well as improved contextual capabilities.
With these upgrades, JAT can now generate a concise summary of a jobseeker’s qualifications, experience, and specific skills for each role.
JAT Version 3: Keeping Job Recommendations Current and Up-to-Date
One limitation that we encountered with OpenAI’s GPT-3.5 was that the data was only relevant until 2021.
This meant that our job recommendations were not up-to-date, which is not very helpful for our career coaches or jobseekers.
We stumbled upon this excellent open source framework called LangChain which was precisely what we needed to address this limitation.
With LangChain, we could augment the results by using a technique called Retrieval-Augmented Generation (or RAG for short).
RAG enables our application to retrieve relevant information from the internet.
We then fed GPT-3.5 with that information and used it to produce easy-to-read summaries as well as links to actual job postings on MyCareersFuture.
The result?
A comprehensive job search experience that provides not only personalised job recommendations but also real-time and up-to-date job listings on MyCareersFuture.
JAT Version X to JARVIS
So what’s next?
We are preparing for the next phase of JAT development in collaboration with GovTech’s MyCareersFuture Team and the Data Science and Artificial Intelligence Division (DSAID) to enhance its functionality.
One significant update planned is the integration of multimodal large language models, or MLLMs (not to be confused with MLMs).
Multimodal LLMs are capable of processing not only text but also other forms of data such as images, audio, and video.
By incorporating these models, JAT will unlock new use cases that were previously impossible with text-only models and possibly help address some of the challenges of the current generation of LLMs.
Our career coaches are like the superheroes of the job market, equipped with the knowledge and skills to guide jobseekers through the challenging job search process.
Who knows, maybe one day JAT will be able to support career coaches the same way JARVIS does for Ironman.
