Case Study - Building systems to Syft through CVs effectively
Syft is a UK based start-up on a mission to reduce risk in the hiring process.
- Client
- Syft
- Year
- Service
- AI product development
Our client
Syft is a UK based start-up on a mission to reduce risk in the hiring process. The business was built to enable recruiters to increase the speed and success rates of recruitment. In doing so, it hopes to allow those hiring to focus on the most enjoyable part of their jobs; meeting people, developing relationships, and seeing their recommendations thrive.
Our challenge
To transform the recruitment process and increase the probability of finding quality candidates for job roles.
As most people in recruitment know, there is a direct correlation between the volume of CVs reviewed and the likelihood of making an exceptional hire. However, this reviewing process can be challenging, especially when it’s now common to receive over a thousand job applications for a single post.
Traditionally, where volumes get high, companies can default to ill-fitting methods to narrow the numbers. Approaches range from random selection (which is both high-risk and detrimental) to meticulous analysis of every CV (which can be time consuming or lead to human error).
But what if we could create a detailed resume review system that did it all, efficiently and effectively?
Scrutinising every word shared
Exploring all links provided
Gathering as much information as possible to thoroughly assess or eliminate each candidate?
Our solution
We worked closely with the Syft team to build a product that addresses three core challenges that recruiters currently face.
-
Parsing and standardising resumes As CVs can feature a diverse range of formats, structures, and wording, it’s previously been difficult to develop a parser (a software component that takes input data such as text and builds a data structure) that is able to handle multiple resume formats.
-
Matching job description to resumes Matching applicants to a certain job description or skill set can be a cumbersome task, especially when you have a wealth of jobs and CVs to manage. No matter how good your filing system may be, keeping abreast of job title changes or transferable skills using traditional methods can be very challenging.
-
Find similar resume (resume to resume matching) There are times in recruitment where you are given the resume of an ideal candidate, in order to find a new hire. Again, using traditional methods, the process of identifying the resume that is most similar to this wish list can be very time-consuming and open to error.
To solve these challenges, the Seeai team utilised the document processing capabilities of Large Language Models (LLMs). We trained an existing LLM to act as a resume and job description parser and taught it to standardise documents. This enables it to convert any document into an analysis-ready format that can be stored and indexed in a database.
Alongside this, we created a pipeline that combined the LLM with Atlas Search (by MongoDB) to further improve the user experience. Utilising existing search engines, we were able to make the most of techniques such as fuzzy search, custom scoring, geospatial-aware search, and speed optimisation.
The results
Not only did we create a solution that solves three core issues for the recruitment community, but we also built it in a way that integrates seamlessly within Syft’s everyday operations.
With each Syft recruiter having their own favourite Applicant Tracking System (ATS) already in place, Syft was keen to build a product that could integrate easily within current workflows (rather than introducing a new one). As such, the Seeai team has wrapped Syft as a Chrome Extension, allowing recruiters to continue with their preferred approaches.
As a result, the client has a tool designed to:
-
Help them gain a competitive advantage
-
Increase their colleagues' success and job satisfaction
-
Encourage colleague engagement in their investment
If you’d like to see the Syft product in action, do explore this demo.
- Full stack development
- LLM integration
- Fewer transactions
- 34%
- Slower transactions
- 10%
- Transaction latency
- 1000ms
- Active nodes
- 3