AI-Powered SaaS Platform for Accelerating The Hiring Process
AutoScreen is an innovative platform for automating and scaling the recruitment process, powered by artificial intelligence. HR managers can create job postings, as well as prepare a set of questions required for potential applicants.
Applicants record video responses online and send them to employers with a pinned CV. Meanwhile, HR managers only have to watch these videos and choose the most relevant candidate for themselves. In their assessments, HR managers use a built-in AI system to check candidates’ video responses.
Who Can Use AutoScreen
HR managers and recruiters who look for new employees, and candidates who look for a new job and usually get prepared for job interviews.
What Did The Client Come To Us With?
Our client wants to order a platform that will be convenient and comfortable for both HR managers and job seekers. They wanted to get such a platform that would significantly reduce the job search and processing experience, knowledge, and skills data of each candidate.
Initially, the client only had a project design in Figma for the platform for candidate selection through interviews and technical documentation.
Project’s First Preparations
Based on the information received from the client, our project manager began developing a basic work plan. We analyzed the project’s scope, calculated the risks, and selected a team based on the experience and skills of the developers. We discussed all the development stages with the client, and he approved our work plan.
In total, a team of 4 people worked on the project. These people include two developers. and two project managers (they changed during development).
Keeping the Contact with a Client
After the initial working plan was approved, we kept the contact with our customer via chat in WhatsApp at the customer’s request. To keep the customer aware of the project’s progress, we discussed our plans and approved daily and weekly reports, and we held calls and video conferences once or twice a week.
Technologies We Chose
We selected technologies based on the client’s request, and the main request was to do everything on Vue.js and implement OpenAI for the functionality of creating automatic questions and chatbots.
Other technologies we also used in AutoScreen were TailwindCSS, pina, axios, tiptap, tinyMCE, Google Cloud, MailGun, Video.js, Wavesurfer.js, Filament, Laravel, Stripe and Plivo.
We implemented tokens into the project to simplify authorisation on a website. These tokens include Sanctum Token, Tolt, Swagger API, Scramble, and Filament.
Description of the Development
The client already had a rough idea of what his project should look like. Therefore, the main design of the site was based on a mockup, fonts, and a palette offered by our client. The rest of the site our team developed, using the initial information provided by the customer. The main page’s final design was also developed by a customer shortly before the AutoScreen release.
The project architecture and database used the proven MySQL. This database management system is characterized by high performance, the ability to handle large amounts of data and high-load queries. In addition, MySQL has a flexible architecture and is suitable for the development of large and small projects and has a variety of administration systems. libraries and frameworks.
During development, the project was divided into front-end and back-end parts in 2 separate repositories, since the connection was via API. For each new feature, back-end code was written first, and then front-end. Of course, the main process of developing and writing codes for the front-end and back-end was carried out in compliance with the principle of three servers: local, staging and production.
In AutoScreen, third-party services Google Cloud, Stripe, and OpenAI were used. Google Cloud was used to implement the Place Autocomplete API. This API is used to install city search when employers fill their company’s address and location. Stripe was used to integrate recurring subscriptions that customers who use the AutoScreen website.
During the development, we used SSH (Security Shell), a cryptographic network protocol that provides a secure connection over unsecured networks. SSH encrypts data transmitted between the client and the server and protects passwords and confidential data.
During development, we always test our work. In general, testing is divided into two areas: manual and automated testing. In this project, we used both methods. Meanwhile, while the QA engineer tested the system using manual testing methods, the developers covered the project code with automatic Unit tests. This helped us minimize errors in the final product.
One of the main challenges in our work was the implementation of a video player for recording interviews, as well as processing recorded video responses, integrating the Plivo API for implementing phone-call interviews. Using various methods described in the Plivo service documentation, plus the help of other developers from our team and our CTO, the development difficulties were resolved.
Improve Users’ Authorisation With Tokens
We have also developed an authorization option via tokens to be able to log in using other services. When candidates (job seekers) or employers register an account, they need information from third-party sites. To do this, the user inserts the token into a third-party site, so the platform can receive the needed data. We introduced this feature in order to remove the constant need to enter a login and password on the site. Instead, the user can update the token once every six months – a virtual key and a set of characters that the user inserts into a third-party site.
In particular, to introduce authorization via token, we used the following technologies:
- Sanctum Token API. This token is used for authorization on third-party sites integrated into the project.
- Tolt. This token is used to create referral links to vacancies on the site. In particular, Tolt helps SaaS startups create and launch affiliate and referral programs.
- Swagger puts API specifications such as OpenAPI, AsyncAPI, and JSON Schema at the heart of its architecture. This is very important for guiding teams throughout the API development and documentation lifecycle.
- Scramble. It’s an OpenAPI documentation generator for Laravel. It automatically generates API documentation for your project without requiring you to manually write PHPDoc annotations.
- Filament is a tool for creating admin panels in Laravel, and also has excellent plugin support.
Refine the project’s efficiency with AI Implementation
One of the features of this project is that we involved artificial intelligence in the work to conduct a conversation with the candidate and evaluate their answers. How was it introduced into the work?
We needed to train this artificial intelligence using machine learning so that it would form comments and questions more correctly. A comment like “answer the question politely” is added to each answer of the person being interviewed. That is, the setting is given to the artificial intelligence, and it accepts it. Then the person’s answer itself is added here, and this answer is processed by the artificial intelligence, and the result is obtained. And this comment is something that we can work with. That is, change it, rephrase it, and then get different answer options.
In addition, the client also wanted to expand the functionality of the platform and involve the phone in calls. Thus, we had two options: use the Trilio service, which you can just Google it separately; it is used for sending SMS, for calls, for video calls, and for calls via phone. Or there was an option for the Plivo service. Since the client needed free speech functionality, after digging into these services, it was discovered that only Plivo has free speech support. Therefore, we decided to integrate Plivo and make calls through it.
Introducing an OpenAI model is another way to improve the project’s efficiency. OpenAI was introduced by our developers at the personal request of our client. This technology is used to generate evaluations of candidates’ video answers and select questions. With the help of given prompts, which include a certain set of static rules, as well as dynamic data (for example: title & job description), requests are sent to OpenAI. In order to fully evaluate the candidate’s video response, it is necessary to carry out a number of the following manipulations: receive a video response from the candidate, save it in storage, convert the video to audio, receive an audio transcription and prepare a prompt. And after then, a request is sent to OpenAI to receive a detailed evaluation according to all available evaluation criteria. On several occasions, there were difficulties with processing the received response from OpenAI, namely with its formatting in different work scenarios, so it was necessary to make changes to the prompt and add different validations.
What Was The Final Solution
Create New Job Listing on AutoScreen
Autoscreen is a project for automatic interviewing, designed for companies that do not have an HR department. At first, employers set up a job name on their account using the create job button. They specify the job title, choose the salary, and the maximum interview duration in the required fields. Then, they select optional parameters, such as job location among the cities that appear in the pop-up window. The next important step – creating and selecting AI-generated questions for the candidate, based on the aforementioned vacancy data. Next, the employer selects the types and formats of answers to the questions for the candidate: video, audio, or text. Then, a user selects evaluation criteria based on characteristics such as leadership, communication, etc. The characteristics that the user selects on the site are also generated by AI. After completing all stages, the vacancy is created on the site and is available to candidates.
Make A First Impression With Video Response
A user who wants to get a job receives an invitation via a link from the desired company to Autoscreen. They register on the site via this link and go through the interview themselves. The system initiates a video call, where a user answers questions prepared in advance. The user’s answers are recorded and converted into text using artificial intelligence. The answers, converted into text, are sent to the recruiter, who analyzes them. After several candidates passed this interview, AI gives assessments to all of their responses. Based on the results, an employer watches and compares the candidates and chooses the most suitable one using artificial intelligence.
How does AI estimate candidates? First, artificial intelligence evaluates a candidate’s CV and the text transcription of the recorded interview. Based on predefined criteria, it gives a score. According to the estimation system, the highest score that can be given to a candidate is 100 points.
Make Phone Calls via AutoScreen
Additionally, AutoScreen contains advanced functionality for employees to pass the interview, which allows them to use a phone. How does this happen? The person receives an invitation link and registers on the AutoScreen website. The user enters their phone number and calls either through Autoscreen or directly to the company’s phone. Before making a phone call, the system checks whether the company’s phone number is in the database. If the phone number isn’t in a system, then it will not be possible to conduct an interview. If the phone number is in the database, the system matches the call with the person’s phone number and account.
What are the next steps? There are two options. The first option is when AI asks the candidate pre-prepared questions in an audio voice. A user answers them, the system records it through a video interview. The system records the interview, converts the audio into text, and gives a score. And the second option – when a person calls, the system asks them one question, and the person answers it. At this moment, the system recognizes the answer, converts it into text, and sends it via Plivo to OpenAI to the artificial intelligence chatbot. And, based on this answer, the artificial intelligence gives its comment or asks the next question itself. Thus, it is as if there is live communication, but with the use of artificial intelligence from Autoscreen.
Project’s Current Status
The platform is in test mode and has been actively underway for 7 months. In particular, our experts are paying special attention to the implementation of the OpenAI model, which generates interview questions and then evaluates the candidate’s answers.
The project’s development and ongoing conversation with a client are still going on. We also look forward to ongoing communication and collaboration throughout the project.
The link: https://autoscreen.ai/