DISCLAIMER: The thought of doing Cowl Letter and even Resume with AI doesn’t clearly begin with me. Lots of people have performed this earlier than (very efficiently) and have constructed web sites and even corporations from the thought. That is only a tutorial on tips on how to construct your personal Cowl Letter AI Generator App utilizing Python and some traces of code. All of the code you’ll say on this weblog submit might be present in my public Github folder. Take pleasure in. 🙂
Pep Guardiola is a (very profitable) Manchester Metropolis soccer coach. Throughout Barcelona’s Leo Messi years, he invented a manner of taking part in soccer generally known as “Tiki-Taka”. Which means as quickly as you obtain the ball, you move the ball, instantly, with out even controlling it. You possibly can move the ball 30–40 occasions earlier than scoring a aim.
Greater than a decade later, we are able to see how the way in which of taking part in soccer made Guardiola and his Barcelona well-known is gone. If you happen to take a look at a Manchester Metropolis match, they take the ball and instantly search for the striker or the winger. You solely want a couple of, vertical passes, instantly on the lookout for the chance. It’s extra predictable, however you do it so many occasions that you’ll finally discover the house to hit the goal.
I believe that the job market has one way or the other gone in the identical route.
Earlier than you had the chance to go to the corporate, hand in your resume, discuss to them, be round them, schedule an interview, and actively discuss to individuals. You’d spend weeks making ready for that journey, sharpening your resume, and reviewing questions and solutions.
For a lot of, this old style technique nonetheless works, and I consider it. In case you have a very good networking alternative, or the appropriate time and place, the handing the resume factor works very effectively. We love the human connection, and it is vitally efficient to truly know somebody.
It is very important take into account that there’s a entire different strategy as effectively. Firms like LinkedIn, Certainly, and even on the whole the web utterly modified the sport. You possibly can ship so many resumes to so many corporations and discover a job out of statistics. AI is altering this sport just a little bit additional. There are lots of AI instruments to tailor your resume for the precise firm, make your resume extra spectacular, or construct the job particular cowl letter. There are certainly many corporations that promote this sort of providers to individuals which are on the lookout for jobs.
Now, consider me, I’ve received nothing in opposition to these corporations, in any respect, however the AI that they’re utilizing it’s probably not “their AI”. What I imply by that’s that for those who use ChatGPT, Gemini, or the tremendous new DeepSeek to do the precise process you’ll very possible not get a worse response than the (paid) instrument that you’re utilizing on their web site. You’re actually paying for the “commodity” of getting a backend API that does what we must do by ChatGPT. And that’s truthful.
Nonetheless, I wish to present you that it’s certainly quite simple and low-cost to make your personal “resume assistant” utilizing Massive Language Fashions. Specifically, I wish to concentrate on cowl letters. You give me your resume and the job description, and I offer you your cowl letter you may copy and paste to LinkedIn, Certainly, or your e mail.
In a single picture, it would appear like this:
Now, Massive Language Fashions (LLMs) are particular AI fashions that produce textual content. Extra particularly, they’re HUGE Machine Studying fashions (even the small ones are very huge).
Which means constructing your personal LLM or coaching one from scratch could be very, very costly. We received’t do something like that. We’ll use a superbly working LLM and we are going to well instruct it to carry out our process. Extra particularly, we are going to do this in Python and utilizing some APIs. Formally, it’s a paid API. Nonetheless, since I began the entire venture (with all of the trial and error course of) I spent lower than 30 cents. You’ll possible spend 4 or 5 cents on it.
Moreover, we are going to make a working net app that can permit you to have your cowl letter in a couple of clicks. It is going to be an effort of a pair hundred traces of code (with areas 🙂).
To inspire you, listed here are screenshots of the ultimate app:
Fairly cool proper? It took me lower than 5 hours to construct the entire thing from scratch. Imagine me: it’s that straightforward. On this weblog submit, we are going to describe, so as:
- The LLM API Technique. This half will assist the reader perceive what LLM Brokers we’re utilizing and the way we’re connecting them.
- The LLM Object. That is the implementation of the LLM API technique above utilizing Python.
- The Net App and outcomes. The LLM Object is then transferred into an internet app utilizing Streamlit. I’ll present you tips on how to entry it and a few outcomes.
I’ll attempt to be as particular as potential so that you’ve the whole lot you should make it your self, but when these items will get too technical, be at liberty to skip to half 3 and simply benefit from the sundown 🙃.
Let’s get began!
1. LLM API Technique
That is the Machine Studying System Design a part of this venture, which I stored extraordinarily gentle, as a result of I wished to maximise the readability of the entire strategy (and since it truthfully didn’t must be extra difficult than that).
We’ll use two APIs:
- A Doc Parser LLM API will learn the Resume and extract all of the significant data. This data can be put in a .json file in order that, in manufacturing, we could have the resume already processed and saved someplace in our reminiscence.
- A canopy letter LLM API. This API will learn the parsed resume (the output of the earlier API) and the job description and it’ll output the Cowl Letter.
Two details:
- What’s the finest LLM for this process? For textual content extraction and summarization, LLama or Gemma are recognized to be a fairly low-cost and environment friendly LLM. As we’re going to use LLama for the summarization process, with a purpose to preserve consistency, we are able to undertake it for the opposite API as effectively. If you wish to use one other mannequin, be at liberty to take action.
- How can we join the APIs? There are a selection of the way you are able to do that. I made a decision to offer it a attempt to Llama API. The documentation is just not precisely intensive, nevertheless it works effectively and it means that you can play with many fashions. You will have to log in, purchase some credit score ($1 is greater than adequate for this process), and save your API key. Be at liberty to modify to a different answer (like Hugging Face or Langchain) for those who really feel prefer it.
Okay, now that we all know what to do, we simply want to truly implement it in Python.
2. LLM Object
The very first thing that we want is the precise LLM prompts. Within the API, the prompts are often handed utilizing a dictionary. As they are often fairly lengthy, and their construction is at all times comparable, it is sensible to retailer them in .json recordsdata. We’ll learn the JSON recordsdata and use them as inputs for the API name.
2.1 LLM Prompts
On this .json file, you should have the mannequin (you may name no matter mannequin you want) and the content material which is the instruction for the LLM. After all, the content material key has a static half, which is the “instruction” and a “dynamic” half, which is the precise enter of the API name. For instance: that is the .json file for the primary API, I known as it resume_parser_api.json:
As you may see from the “content material” there may be the static name:
“You’re a resume parser. You’ll extract data from this resume and put them in a .json file. The keys of your dictionary can be first_name, last_name, location, work_experience, school_experience, expertise. In choosing the data, preserve monitor of essentially the most insightful.”
The keys I wish to extract out of my “.json” recordsdata are:
[first_name, last_name, location, work_experience, school_experience, skills]Be at liberty so as to add something extra data that you simply wish to be “extracted” out of your resume, however keep in mind that that is stuff that ought to matter solely to your cowl letter. The precise resume can be added after this textual content to kind the total name/instruction. Extra on that later.
The order instruction is the cover_letter_api.json:
Now the instruction is that this one:
“You’re an skilled in job looking and a canopy letter author. Given a resume json file, the job description, and the date, write a canopy letter for this candidate. Be persuasive {and professional}. Resume JSON: {resume_json} ; Job Description: {job_description}, Date: {date}”
As you may see, there are three placeholders: “Resume_json”, “job_description” and “date”. As earlier than, these placeholders will then get replaced with the right data to kind the total immediate.
2.2 constants.py
I made a really small constants.py file with the trail of the 2 .json immediate recordsdata and the API that it’s important to generate from LLamaApi (or actually no matter API you’re utilizing). Modify this if you wish to run the file regionally.
2.3 file_loader.py
This file is a group of “loaders” to your resume. Boring stuff however vital.
2.4 cover_letter.py
The entire implementation of the LLM Technique might be discovered on this object that I known as CoverLetterAI. There it’s:
I spent fairly a while attempting to make the whole lot modular and simple to learn. I additionally made lots of feedback to all of the capabilities so you may see precisely what does what. How can we use this beast?
So the entire code runs in 5 easy traces. Like this:
from cover_letter import CoverLetterAI
cover_letter_AI = CoverLetterAI()
cover_letter_AI.read_candidate_data('path_to_your_resume_file')
cover_letter_AI.profile_candidate()
cover_letter_AI.add_job_description('Insert job description')
cover_letter_AI.write_cover_letter()So so as:
- You name the CoverLetterAI object. It is going to be the star of the present
- You give me the trail to your resume. It may be PDF or Phrase and I learn your data and retailer them in a variable.
- You name profile_candidate(), and I run my first LLM. This course of the candidate phrase information and creates the .json file we are going to use for the second LLM
- You give me the job_description and also you add it to the system. Saved.
- You name write_cover_letter() and I run my second LLM that generates, given the job description and the resume .json file, the quilt letter
3. Net App and Outcomes
So that’s actually it. You noticed all of the technical particulars of this weblog submit within the earlier paragraphs.
Simply to be further fancy and present you that it really works, I additionally made it an internet app, the place you may simply add your resume, add your job description and click on generate cowl letter. That is the hyperlink and that is the code.
Now, the cowl letters which are generated are scary good.
This can be a random one:
February 1, 2025
Hiring Supervisor,
[Company I am intentionally blurring]I’m thrilled to use for the Distinguished AI Engineer place at [Company I am intentionally blurring], the place I can leverage my ardour for constructing accountable and scalable AI techniques to revolutionize the banking business. As a seasoned machine studying engineer and researcher with a powerful background in physics and engineering, I’m assured that my expertise and expertise align with the necessities of this position.
With a Ph.D. in Aerospace Engineering and Engineering Mechanics from the College of Cincinnati and a Grasp’s diploma in Physics of Advanced Methods and Large Knowledge from the College of Rome Tor Vergata, I possess a singular mix of theoretical and sensible data. My expertise in growing and deploying AI fashions, designing and implementing machine studying algorithms, and dealing with massive datasets has geared up me with the talents to drive innovation in AI engineering.
As a Analysis and Educating Assistant on the College of Cincinnati, I utilized surrogate fashions to detect and classify cracks in pipes, reaching a 14% enchancment in injury detection experiments. I additionally developed surrogate fashions utilizing deep studying algorithms to speed up Finite Factor Strategies (FEM) simulations, leading to a 1M-fold discount in computational time. My expertise in instructing and creating programs in sign processing and picture processing for teenagers inquisitive about AI has honed my means to speak complicated ideas successfully.
In my earlier roles as a Machine Studying Engineer at Gen 9, Inc., Apex Microdevices, and Accenture, I’ve efficiently designed, developed, and deployed AI-powered options, together with configuring mmWave radar and Jetson units for information assortment, implementing state-of-the-art level cloud algorithms, and main the FastMRI venture to speed up MRI scan occasions. My experience in programming languages comparable to Python, TensorFlow, PyTorch, and MATLAB, in addition to my expertise with cloud platforms like AWS, Docker, and Kubernetes, has enabled me to develop and deploy scalable AI options.
I’m significantly drawn to [Company I am intentionally blurring] dedication to creating accountable and dependable AI techniques that prioritize buyer expertise and ease. My ardour for staying abreast of the most recent AI analysis and my means to judiciously apply novel methods in manufacturing align with the corporate’s imaginative and prescient. I’m excited in regards to the alternative to work with a cross-functional workforce of engineers, analysis scientists, and product managers to ship AI-powered merchandise that rework how [Company I am intentionally blurring] serves its clients.
Along with my technical expertise and expertise, I possess glorious communication and presentation expertise, which have been demonstrated by my technical writing expertise at In direction of Knowledge Science, the place I’ve written complete articles on machine studying and information science, reaching a broad viewers of 50k+ month-to-month viewers.
Thanks for contemplating my utility. I’m keen to debate how my expertise and expertise can contribute to the success of the [Company I am intentionally blurring] and [Company I am intentionally blurring]’s mission to convey humanity and ease to banking by AI. I’m assured that my ardour for AI, my technical experience, and my means to work collaboratively will make me a precious asset to your workforce.
Sincerely,
Piero Paialunga
They appear identical to I might write them for a selected job description. That being mentioned, in 2025, you should watch out as a result of hiring managers do know that you’re utilizing AI to put in writing them and the “pc tone” is fairly straightforward to identify (e.g. phrases like “keen” are very ChatGPT-ish lol). For that reason, I’d prefer to say to use these instruments correctly. Positive, you may construct your “template” with them, however you should definitely add your private contact, in any other case your cowl letter can be precisely like the opposite hundreds of canopy letters that the opposite candidates are sending in.
That is the code to construct the net app.
4. Conclusions
On this weblog article, we found tips on how to use LLM to transform your resume and job description into a selected cowl letter. These are the factors we touched:
- The usage of AI in job looking. Within the first chapter we mentioned how job looking is now utterly revolutionized by AI.
- Massive Language Fashions concept. It is very important design the LLM APIs correctly. We did that within the second paragraph
- LLM API implementation. We used Python to implement the LLM APIs organically and effectively
- The Net App. We used streamlit to construct a Net App API to show the facility of this strategy.
- Limits of this strategy. I believe that AI generated cowl letters are certainly superb. They’re on level, skilled and effectively crafted. Nonetheless, if everybody begins utilizing AI to construct cowl letters, all of them actually look the identical, or not less than all of them have the identical tone, which isn’t nice. One thing to consider.
5. References and different sensible implementations
I really feel that’s simply truthful to say lots of sensible folks that have had this concept earlier than me and have made this public and out there for anybody. That is just a few of them I discovered on-line.
Cowl Letter Craft by Balaji Kesavan is a Streamlit app that implements a really comparable concept of crafting the quilt letter utilizing AI. What we do totally different from that app is that we extract the resume straight from the phrase or PDF, whereas his app requires copy-pasteing. That being mentioned, I believe the man is extremely gifted and really inventive and I like to recommend giving a glance to his portoflio.
Randy Pettus has a comparable concept as effectively. The distinction between his strategy and the one proposed on this tutorial is that he’s very particular within the data, asking questions like “present hiring supervisor” and the temperature of the mannequin. It’s very attention-grabbing (and good) that you would be able to clearly see the way in which he’s considering of Cowl Letters to information the AI to construct it the way in which he likes them. Extremely beneficial.
Juan Esteban Cepeda does an excellent job in his app as effectively. You too can inform that he was engaged on making it larger than a easy streamlit add as a result of he added the hyperlink to his firm and a bunch of opinions by customers. Nice job and nice hustle. 🙂
6. About me!
Thanks once more to your time. It means rather a lot ❤
My identify is Piero Paialunga and I’m this man right here:
I’m a Ph.D. candidate on the College of Cincinnati Aerospace Engineering Division and a Machine Studying Engineer for Gen 9. I discuss AI, and Machine Studying in my weblog posts and on Linkedin. If you happen to preferred the article and wish to know extra about machine studying and observe my research you may:
A. Comply with me on Linkedin, the place I publish all my tales
B. Subscribe to my publication. It’s going to preserve you up to date about new tales and provide the likelihood to textual content me to obtain all of the corrections or doubts you’ll have.
C. Turn into a referred member, so that you received’t have any “most variety of tales for the month” and you may learn no matter I (and hundreds of different Machine Studying and Knowledge Science prime writers) write in regards to the latest expertise out there.
D. Need to work with me? Verify my charges and tasks on Upwork!
If you wish to ask me questions or begin a collaboration, depart a message right here or on Linkedin:
