#GPT
#OpenAI
#Automation
AI Cover Letter Creator
AI Cover Letter Creator
AI Cover Letter Creator
AI Cover Letter Creator
AI Cover Letter Creator
AI Cover Letter Creator
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Cover Letter Creator was my first attempt at building a useful Generative Pre-trained Transformer (GPT) to solve a minor but annoying problem I faced. It is a live custom GPT posted to OpenAI's GPT store.
Feb 2024 - Mar 2024
Timeline
Builder
Role
Feb 2024 - Mar 2024
Timeline
Builder
Role
Background
Background
In the realm of technology, artificial intelligence (AI) and machine learning have emerged as game-changers, particularly for rules-based tasks. Building applications that leverage AI model capabilities is a space ripe with opportunity.
This was a minor project that aimed to automate writing a personalised, job-relevant cover letter. It is a relatively insignificant problem I faced, but still a time-consuming 'box-ticker' task for many job applicants. I believed that the characteristics and use-case of a cover letter allow it to be an excellent target for an AI-based solution.
In the realm of technology, artificial intelligence (AI) and machine learning have emerged as game-changers, particularly for rules-based tasks. Building applications that leverage AI model capabilities is a space ripe with opportunity.
This was a minor project that aimed to automate writing a personalised, job-relevant cover letter. It is a relatively insignificant problem I faced, but still a time-consuming 'box-ticker' task for many job applicants. I believed that the characteristics and use-case of a cover letter allow it to be an excellent target for an AI-based solution.
Goals
Goals
1
Build a Cover Letter Generator that I'd love to use.
From what I had learnt so far, solving a problem that is my own is a great way to ensure I'm solving a real problem. For a project of this scale, I'd was aiming to build a bot that can serve my own use-case well.
1
Build a Cover Letter Generator that I'd love to use.
From what I had learnt so far, solving a problem that is my own is a great way to ensure I'm solving a real problem. For a project of this scale, I'd was aiming to build a bot that can serve my own use-case well.
1
Build a Cover Letter Generator that I'd love to use.
From what I had learnt so far, solving a problem that is my own is a great way to ensure I'm solving a real problem. For a project of this scale, I'd was aiming to build a bot that can serve my own use-case well.
2
Understand how OpenAI's GPT Store and platform functions.
There had been a lot of hype around OpenAI's GPT store and the potential for custom GPT based applications ('AI Wrappers'). I wanted to understand if this platform was right for me to build some larger projects. Gaining some experience developing 'Layer 2' AI applications was also a useful by-product.
2
Understand how OpenAI's GPT Store and platform functions.
There had been a lot of hype around OpenAI's GPT store and the potential for custom GPT based applications ('AI Wrappers'). I wanted to understand if this platform was right for me to build some larger projects. Gaining some experience developing 'Layer 2' AI applications was also a useful by-product.
2
Understand how OpenAI's GPT Store and platform functions.
There had been a lot of hype around OpenAI's GPT store and the potential for custom GPT based applications ('AI Wrappers'). I wanted to understand if this platform was right for me to build some larger projects. Gaining some experience developing 'Layer 2' AI applications was also a useful by-product.
3
Lower the barrier to high quality AI outputs.
The ability to effectively prompt LLM based AI models was key to obtaining useful outputs. I aimed to ensure my GPT took that variable out of the equation by providing the user with a 'hand-held' prompting experience.
3
Lower the barrier to high quality AI outputs.
The ability to effectively prompt LLM based AI models was key to obtaining useful outputs. I aimed to ensure my GPT took that variable out of the equation by providing the user with a 'hand-held' prompting experience.
3
Lower the barrier to high quality AI outputs.
The ability to effectively prompt LLM based AI models was key to obtaining useful outputs. I aimed to ensure my GPT took that variable out of the equation by providing the user with a 'hand-held' prompting experience.
Process
Process
1
Test current Cover Letter GPTs to see if my problem has already been solved.
I began this project by testing existing cover letter GPTs on OpenAI's GPT store. I noticed several tools had quite high usage numbers (>25k chats) which reinforced my belief that this was a real problem encountered by others. Interestingly, the most popular GPTs in the store produced outputs that I judged to be poor and 'generic'. Despite being easy to use, the existing solutions hadn't solved my problem yet.
1
Test current Cover Letter GPTs to see if my problem has already been solved.
I began this project by testing existing cover letter GPTs on OpenAI's GPT store. I noticed several tools had quite high usage numbers (>25k chats) which reinforced my belief that this was a real problem encountered by others. Interestingly, the most popular GPTs in the store produced outputs that I judged to be poor and 'generic'. Despite being easy to use, the existing solutions hadn't solved my problem yet.
1
Test current Cover Letter GPTs to see if my problem has already been solved.
I began this project by testing existing cover letter GPTs on OpenAI's GPT store. I noticed several tools had quite high usage numbers (>25k chats) which reinforced my belief that this was a real problem encountered by others. Interestingly, the most popular GPTs in the store produced outputs that I judged to be poor and 'generic'. Despite being easy to use, the existing solutions hadn't solved my problem yet.
2
Design a way to attain better outputs.
Despite great UX's, I believed existing model poor performance was in part due to the frictionless approach other builders took. The popular models sourced all user data from a single file upload and all role data from a single webpage. OCR reading of resume documents and scraping an entire webpage for role data creates too much variance in model output. If a resume was not easily machine readable due to formatting, or a role application page was flooded with garbage 'corporate speak' text, I believed the GPT would struggle with extracting the data required to produce letters that are 'ready'. My idea was to enable the user to adopt a structured data input approach. Combined with some back-end prompting I believed this would reduce uncertainty and errors at the data sourcing, resulting in better overall outputs.
2
Design a way to attain better outputs.
Despite great UX's, I believed existing model poor performance was in part due to the frictionless approach other builders took. The popular models sourced all user data from a single file upload and all role data from a single webpage. OCR reading of resume documents and scraping an entire webpage for role data creates too much variance in model output. If a resume was not easily machine readable due to formatting, or a role application page was flooded with garbage 'corporate speak' text, I believed the GPT would struggle with extracting the data required to produce letters that are 'ready'. My idea was to enable the user to adopt a structured data input approach. Combined with some back-end prompting I believed this would reduce uncertainty and errors at the data sourcing, resulting in better overall outputs.
2
Design a way to attain better outputs.
Despite great UX's, I believed existing model poor performance was in part due to the frictionless approach other builders took. The popular models sourced all user data from a single file upload and all role data from a single webpage. OCR reading of resume documents and scraping an entire webpage for role data creates too much variance in model output. If a resume was not easily machine readable due to formatting, or a role application page was flooded with garbage 'corporate speak' text, I believed the GPT would struggle with extracting the data required to produce letters that are 'ready'. My idea was to enable the user to adopt a structured data input approach. Combined with some back-end prompting I believed this would reduce uncertainty and errors at the data sourcing, resulting in better overall outputs.
3
Build and Test the GPT.
I built the GPT using Open Ai's builder platform. This method was incredibly easy and did not require setting up a seperate API or interface. To provide users with a consistent, structured input option, I built out a prompt skeleton that was printed out to the user after their request to write a cover letter. This skeleton assisted users to provide all required data to the GPT, and also assisted the GPT to personalise and build a better overall letter. After this, I tested the bot with my resume data and several roles. It required lots of tweaking in terms of language and tone, but the data sourcing issue had been solved. To complete the build I added an option for users to input their existing desired structure template to the GPT after some feedback from close friends and other testers.
3
Build and Test the GPT.
I built the GPT using Open Ai's builder platform. This method was incredibly easy and did not require setting up a seperate API or interface. To provide users with a consistent, structured input option, I built out a prompt skeleton that was printed out to the user after their request to write a cover letter. This skeleton assisted users to provide all required data to the GPT, and also assisted the GPT to personalise and build a better overall letter. After this, I tested the bot with my resume data and several roles. It required lots of tweaking in terms of language and tone, but the data sourcing issue had been solved. To complete the build I added an option for users to input their existing desired structure template to the GPT after some feedback from close friends and other testers.
3
Build and Test the GPT.
I built the GPT using Open Ai's builder platform. This method was incredibly easy and did not require setting up a seperate API or interface. To provide users with a consistent, structured input option, I built out a prompt skeleton that was printed out to the user after their request to write a cover letter. This skeleton assisted users to provide all required data to the GPT, and also assisted the GPT to personalise and build a better overall letter. After this, I tested the bot with my resume data and several roles. It required lots of tweaking in terms of language and tone, but the data sourcing issue had been solved. To complete the build I added an option for users to input their existing desired structure template to the GPT after some feedback from close friends and other testers.
4
Evaluate whether it is worth pursuing this opportunity further.
Having built the GPT, I needed to decide if I wanted to take this idea further or leave it in it's current state. Whilst I was quite happy with the performance, I didn't think it was enough of a differentiator to take further via a GPT wrapper on it's own website. Additionally, I realised I didn't think of a strategy or real plan to actually distribute the GPT to more people. In retrospect, a TikTok account that reviewed GPTs for effectiveness could've been a great way to organically promote the product.
4
Evaluate whether it is worth pursuing this opportunity further.
Having built the GPT, I needed to decide if I wanted to take this idea further or leave it in it's current state. Whilst I was quite happy with the performance, I didn't think it was enough of a differentiator to take further via a GPT wrapper on it's own website. Additionally, I realised I didn't think of a strategy or real plan to actually distribute the GPT to more people. In retrospect, a TikTok account that reviewed GPTs for effectiveness could've been a great way to organically promote the product.
4
Evaluate whether it is worth pursuing this opportunity further.
Having built the GPT, I needed to decide if I wanted to take this idea further or leave it in it's current state. Whilst I was quite happy with the performance, I didn't think it was enough of a differentiator to take further via a GPT wrapper on it's own website. Additionally, I realised I didn't think of a strategy or real plan to actually distribute the GPT to more people. In retrospect, a TikTok account that reviewed GPTs for effectiveness could've been a great way to organically promote the product.
Tools
Tools
Open AI
Microsoft Office
Final Thoughts
Whilst I achieved my goal of building a Cover Letter Generator that serves my personal use-case, the output is not ready to send to a recruiter after generation. (Output still isn't good enough without additional prompting)
I gained some useful experience in using Open AI's developer GPT developer platform, but realised some critical pitfalls of the platform in doing so.
Despite producing better results then the most popular cover letter GPT on Open AI's store, a lack of high-level differentiation and a distribution/marketing strategy will hinder this tool's success.
Overall, I learnt several surprisingly useful things by building this minor tool that I will take into my next AI application build.
