If you want to innovate, you have to iterate!
It has been 30 days since I joined Beamery, an HR SaaS B2B company. I started a new role as a Lead Data Scientist in the Edge team — a group of applied data scientists and AI practitioners. Before this, I worked at CogX as a tech lead for the data science team, where I was involved in applying different AI algorithms on the CogX app.
Here in this blog post, I aim to focus on my first 30 days at Beamery as a data scientist. I want to tell you about Beamery, the Edge team, the data science projects, and my focus for the next couple of months.
My agenda for the first month:
My main aim for the first month was simple: get to know my new colleagues and learn about ongoing AI projects to see how I can collaborate and add value to the team. I came up with the following steps :
#1 Get to know my teammates and their projects:
The Data Science team at Beamery is called “Edge AI”, which is at the core of Beamery’s vision for the future. The team is serviced by the Beamery Talent Graph, which is a knowledge graph of 20 billion+ facts modeling the world of core talent and recruitment concepts; This means all ML models benefit from the massive amount of connected concepts and facts, and then these models are made available to Beamery’s customers to apply to their own data.
I started to set 1:1 meetings with the team to know the projects more in detail. In my meetings with the team I was trying to find answers to these questions:
- What are the biggest challenges in the project and why?
- What tools/software/frameworks have been used?
- How is the project linked to the other engineering teams at Beamery?
- How can I tackle the challenge?
Having spoken to everyone, it became clear that there are many ongoing data science projects within the Edge team including (1) Predicting employee’s move (to a new job or to new seniority level), (2) Skill prediction, (3) Matching, (4) Candidate Engagement Scoring, (5) Workforce Planning, (6) Career trajectory prediction, (7) Diversity, ethnicity and age inference. I hope to write more in detail about each of them in the future.
#2 Getting to know the “Beamery Knowledge Graph”
After my initial meetings with colleagues, the importance of the knowledge graph became obvious — all of the data points for all AI projects come from the knowledge graph. I needed to get familiar with it. Beamery’s knowledge graph, also known as Beamery’s semantic network or talent knowledge graph, represents a network of HR entities such as the social domain of people, their skills, work experience, companies, education, and location. It illustrates the relationship between these entities too.
The knowledge graph has been developed based on two main data sources : (1) Enterprise HRIS, HRMS, and other company data sources that provide a holistic picture of an organisation when stitched together as knowledge. (2) Free and openly available data such as social media, news sources, company information, administrative data, and other linked datasets such as DBpedia.
So underpinning all AI works at Beamery is the talent knowledge graph, which is quite adaptive and dynamic. If you like to read more about Beamery’s knowledge graph, have a look at blog posts by my smart colleagues Henri Egle Sorots and Kasper Piskorski . Well worth a read!
#3 Adjusting my focus based on value complexity matrix
So after a few meetings and discussions with the team and our head of AI, Ahmad Assaf, I decided to focus on a project with high value and low complexity (Fig 3). I chose to take on “Skill prediction and the matching algorithm” and actively collaborate on developing the next iterations.
The current version of the algorithm, at a very high level, is based on a deep graph convolutional network, where all the inputs to the model are derived from the knowledge graph, and the model has been set to infer skills for role titles. Role title representation is one of the byproducts of this model, where we can map all role titles into a numerical vector in the vector space model through an iterative refinement process of the deep neural network. My short-term objectives are to (1) improve the role representation vector so we will have a more precise skill prediction and (2) establish an evaluation algorithm to be able to do a comparative analysis over different iterations of our algorithms. By the time of writing this blogpost I came up with a PoC on role title representation, and I’m in the scaling-up process. I’m working closely with my great colleague Kaan Karakeben on this.
#4 Defining my longer-term objectives
The application of AI in HR has been one of my favourite topics for years, and the story started when my husband, launched his own AI startup — Cyra- AI-based recruiting assistant. We had been always discussing Cyra’s different challenges and problems, and the way that AI can address them. Since then, one of my favourite topics in the HR domain was predicting professional career trajectories. It basically means predicting your next job! So I discussed my interest in this project with the team, and then chose this exciting project as my main project for the next couple of months. We do quarterly planning at Edge and this project has been set on my radar for Q1 and Q2. I might write about our quarterly planning at some point in the future.
#5 Looking for collaborators
To develop an AI model, the last thing you want is to reinvent the wheel. So, I started to look into what has been done before at Beamery regarding the career trajectory project, and I realised one of our teams in the US has worked on similar problems before, but with slightly different use cases. They have aimed to predict the next job for a candidate inside the current candidate company, whereas my objective is to predict the next job inside/outside the current candidate company. I discussed this with the US team, and we are set to meet more regularly to make sure we are on the same page and doing the work in a way that can benefit both sides. By the time that I am writing this blog post, I am wrapping up my literature reviews and will be slowly moving to the next step to build our first career trajectory MVP.
These were my 5 main steps within my first 30 days at Beamery as a lead data scientist. I believe AI plays an important role to make the HR world easier, making the hiring process faster and much more efficient than it was before. I will share my research outcomes and findings here to let you know how our projects progress.
🇺🇦These last few weeks at Beamery have been great, but it’s also been a very sad time to watch the war in Ukraine unfold. So, I wanted to take a moment to show my solidarity with Ukraine in this article.
Interested in joining our Engineering, Product & Design Team?
We’re looking for Data Scientists, Software Engineers Front & Back Mid/Senior/Director, SRE Platform Engineers, Engineering Managers, Tech Leads, Product Operations/Managers/Designers across all levels and even more roles across London, Remote, USA & Berlin! to join us — apply here!