Case study – AI used for the systematic generation of job advertisements
Are you already using the incredible possibilities of artificial intelligence?
If you are not yet able to really categorize what you can systematically use artificial intelligence for, this case study will show you a good example.
Intro
Together with a team of 10 developers, I built a SaaS product for a client whose service consists of generating job ads from a short input prompt, which are displayed as ads on Facebook and Instagram. Potential applicants are taken from the ad to a landing page, where they enter their contact details and can thus be seen in the system and contacted.
Until now, it took a lot of effort for head hunters or companies to place job advertisements on Facebook and link them to a dedicated landing page.
This usually fell under the jurisdiction of online marketing agencies and had its price.
Background
For many, the topic of AI is difficult to grasp. We are currently experiencing an arms race between the titans in the development of artificial intelligence, where OpenAI, Google, Meta, Amazon, Tesla, Baidoo, etc. are vying for supremacy.
I like to compare the situation with the gold rush in the USA in the 19th century.
Some gold prospectors were lucky and made a fortune, while many struggled to survive. As soon as the mines no longer yielded a profit, the prospectors’ source of income dried up and they were forced to move on.
However, brands such as Levi’s, which produced tools and clothing for the prospectors and generated business from this, proved to be more stable.
They were not directly dependent on the mining sites and could operate in different places where their products were in demand.
Similarly, I focus on the application of the new AI tools and like to leave the development of the models to the big players, even though it is not that complicated to develop AI models yourself.
Content generation with AI
Everyone today is probably familiar with chatGPT as an application for individual users. You have probably already gained experience with various AI applications and generated something useful while drastically reducing the previous effort.
As a solution architect and developer, it was obvious that these new AI tools could not just be used to generate individual fragments for personal use. The major AI providers naturally allow access to their systems via APIs.
This works in such a way that an application logs in to the AI system, e.g. chatGPT or OpenAI, via authentication. The application can then send as many requests to the API of the AI system as it needs and receives a response to each request, which can then be used in the application.
We therefore developed a query prompt for chatGPT, into which the user’s job advertisement details are integrated.
The user’s entry can be as simple as “I am looking for a master baker in Zurich” or it can contain a full classic job description.
To do this, we defined which values we need as a response and in which form they should be returned. In the technical environment, JSON is a good choice for the data structure of the response.
As we have no control over the user’s input in the prompt, it is very important to first ensure that chatGPT interprets the job advertisement correctly. We therefore use the prompt to exclude everything that violates the META rules for ads on Facebook and Instagram.
We also search the prompt for specific definitions such as the job title, location, benefits and job requirements.
Prompt Engineering
You have probably already had some experience with prompt engineering and know that the more meaningful information it is given about the context of the query, the better the results of an AI query will be. To generate text, you can tell the AI how the style of the wording should be kept, for example, or you can specify that the text should be written from the perspective of an online marketer, for example.
For a job title such as “master baker”, the AI happily provides the classic requirements and common benefits in response to a nice question. Information such as geodata for a location is also provided courteously.
Creation of image material with AI
For the ads on Facebook, we also need image material. ChatGPT will be happy to provide us with a prompt for generating image material with DALL-E-3, the AI image generator from OpenAI, on request.
So as soon as we receive the answer from chatGPT, we send another request to the DALL-E-3 API for an image matching the job description, for which we use the prompt provided by chatGPT to generate the image.
Voila, we now have all the data including image material for the ad on Facebook and can transmit it to META via the API.
Of course, we also asked chatGPT to generate the dynamic content for the landing page. The ads receive the correct URL to the landing page as a target, so the right content can be displayed there.
We simplify a process and drastically reduce the manual effort so that the user can focus on the actual problem of precisely defining the job.
As a programmer, I am used to formulating instructions to computers in a clear and structured way. When dealing with AI, it is also crucial to formulate precisely, clearly and in a structured way what you expect as an answer.
Conclusion
This description uses a simple case study to demonstrate the power of AI systems.
We use OpenAI in the SaaS application because OpenAI dominated the market at the time of development. The development of the various providers is progressing at breakneck speed, so that it is of course necessary to analyze exactly which of the AI providers is ahead for a specific topic.
It also depends on the business idea of the SaaS company’s product which artifacts need to be created. In this example, we needed texts and images, but today we could always find an AI provider where we can generate presentations or audio files. Soon this will probably also be the case for videos.
One experience we have had is that the generation takes a moment, depending on the prompt and AI system.
Accordingly, asynchronous handling of request and response must be taken into account in the system architecture.
I hope I have been able to inspire you with ideas about where and how you can use AI with a systematic and automated approach in your applications.
If you have any questions or would like advice, please contact me.