From Resume to Cover Letter Using AI and LLM, with Python and Streamlit

DISCLAIMER: The thought of doing Cowl Letter and even Resume with AI doesn’t clearly begin with me. Lots of people have completed this earlier than (very efficiently) and have constructed web sites and even corporations from the thought. That is only a tutorial on learn how to construct your individual Cowl Letter AI Generator App utilizing Python and some strains of code. All of the code you’ll say on this weblog publish 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 enjoying soccer generally known as “Tiki-Taka”. Because of this as quickly as you obtain the ball, you move the ball, instantly, with out even controlling it. You’ll be able to move the ball 30–40 instances earlier than scoring a aim.

Greater than a decade later, we will see how the best way of enjoying soccer made Guardiola and his Barcelona well-known is gone. In case you 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 searching for the chance. It’s extra predictable, however you do it so many instances that you’ll ultimately discover the house to hit the goal.

I feel that the job market has by some means gone in the identical route. 

Earlier than you had the chance to go to the corporate, hand in your resume, speak to them, be round them, schedule an interview, and actively speak 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 school technique nonetheless works, and I imagine it. In case you have networking alternative, or the proper time and place, the handing the resume factor works very nicely. We love the human connection, and it is vitally efficient to truly know somebody. 

You will need to think about that there’s a complete different method as nicely. Firms like LinkedIn, Certainly, and even generally the web utterly modified the sport. You’ll be able to ship so many resumes to so many corporations and discover a job out of statistics. AI is altering this sport a bit bit additional. There are numerous AI instruments to tailor your resume for the particular 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 might be searching for jobs.

Now, imagine me, I’ve obtained nothing in opposition to these corporations, in any respect, however the AI that they’re utilizing it’s not likely “their AI”. What I imply by that’s that when you use ChatGPT, Gemini, or the tremendous new DeepSeek to do the precise process you’ll very seemingly not get a worse response than the (paid) instrument that you’re utilizing on their web site. You might be actually paying for the “commodity” of getting a backend API that does what we must do via ChatGPT. And that’s honest. 

Nonetheless, I need to present you that it’s certainly quite simple and low-cost to make your individual “resume assistant” utilizing Massive Language Fashions. Specifically, I need to concentrate on cowl letters. You give me your resume and the job description, and I offer you your cowl letter you possibly can copy and paste to LinkedIn, Certainly, or your electronic mail.

In a single picture, it is going to 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 massive). 

Because of this constructing your individual LLM or coaching one from scratch could be very, very costly. We received’t do something like that. We are going to use a wonderfully working LLM and we’ll neatly instruct it to carry out our process. Extra particularly, we’ll try 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 seemingly spend 4 or 5 cents on it.  

Moreover, we’ll make a working net app that can mean you can have your cowl letter in a couple of clicks. It will likely be an effort of a pair hundred strains of code (with areas 🙂).

To encourage you, listed below are screenshots of the ultimate app:

Fairly cool proper? It took me lower than 5 hours to construct the entire thing from scratch. Consider me: it’s that straightforward. On this weblog publish, we’ll describe, so as:

1. The LLM API Technique. This half will assist the reader perceive what LLM Brokers we’re utilizing and the way we’re connecting them.
2. The LLM Object. That is the implementation of the LLM API technique above utilizing Python. 
3. The Net App and outcomes. The LLM Object is then transferred into an internet app utilizing Streamlit. I’ll present you learn how to entry it and a few outcomes. 

I’ll attempt to be as particular as attainable so that you’ve every part you should make it your self, but when these items will get too technical, be happy 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 mild, as a result of I needed to maximise the readability of the entire method (and since it truthfully didn’t have to be extra difficult than that).

We are going to use two APIs:

1. A Doc Parser LLM API will learn the Resume and extract all of the significant info. This info will probably be put in a .json file in order that, in manufacturing, we can have the resume already processed and saved someplace in our reminiscence.
2. 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:

1. What’s the finest LLM for this process? For textual content extraction and summarization, LLama or Gemma are identified to be a fairly low-cost and environment friendly LLM. As we’re going to use LLama for the summarization process, with a view to hold consistency, we will undertake it for the opposite API as nicely. If you wish to use one other mannequin, be happy to take action.
2. How can we join the APIs? There are a selection of the way you are able to do that. I made a decision to provide it a attempt to Llama API. The documentation will not be precisely in depth, but it surely works nicely and it lets you play with many fashions. You will want 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 resolution (like Hugging Face or Langchain) when you 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 normally handed utilizing a dictionary. As they are often fairly lengthy, and their construction is at all times comparable, it is smart to retailer them in .json information. We are going to learn the JSON information and use them as inputs for the API name. 

2.1 LLM Prompts

On this .json file, you’ll have the mannequin (you possibly can 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 particular 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 possibly can see from the “content material” there’s the static name:

“You’re a resume parser. You’ll extract info from this resume and put them in a .json file. The keys of your dictionary will probably be first_name, last_name, location, work_experience, school_experience, expertise. In deciding on the knowledge, hold observe of essentially the most insightful.”

The keys I need to extract out of my “.json” information are:

[first_name, last_name, location, work_experience, school_experience, skills]

Be at liberty so as to add something extra info that you just need to be “extracted” out of your resume, however do not forget that that is stuff that ought to matter solely in your cowl letter. The precise resume will probably 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 might be 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 possibly can see, there are three placeholders: “Resume_json”, “job_description” and “date”. As earlier than, these placeholders will then get replaced with the right info 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 information and the API that you must generate from LLamaApi (or actually no matter API you might be utilizing). Modify this if you wish to run the file regionally. 

2.3 file_loader.py

This file is a group of “loaders” in your resume. Boring stuff however essential. 

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 making an attempt to make every part modular and simple to learn. I additionally made numerous feedback to all of the features so you possibly can see precisely what does what. How can we use this beast?

So the entire code runs in 5 easy strains. 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:

1. You name the CoverLetterAI object. It will likely be the star of the present
2. You give me the trail to your resume. It may be PDF or Phrase and I learn your info and retailer them in a variable.
3. You name profile_candidate(), and I run my first LLM. This course of the candidate phrase information and creates the .json file we’ll use for the second LLM 
4. You give me the job_description and also you add it to the system. Saved.
5. You name write_cover_letter() and I run my second LLM that generates, given the job description and the resume .json file, the duvet letter

3. Net App and Outcomes

So that’s actually it. You noticed all of the technical particulars of this weblog publish within the earlier paragraphs.

Simply to be additional fancy and present you that it really works, I additionally made it an internet app, the place you possibly can simply add your resume, add your job description and click on generate cowl letter. That is the link and that is the code.

Now, the cowl letters which might be generated are scary good.

It is 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 methods to revolutionize the banking trade. As a seasoned machine studying engineer and researcher with a robust 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 Techniques and Huge Knowledge from the College of Rome Tor Vergata, I possess a novel mix of theoretical and sensible information. My expertise in growing and deploying AI fashions, designing and implementing machine studying algorithms, and dealing with giant datasets has outfitted me with the abilities 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 Aspect 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 enthusiastic about AI has honed my skill to speak advanced 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 gadgets for knowledge assortment, implementing state-of-the-art level cloud algorithms, and main the FastMRI venture to speed up MRI scan instances. My experience in programming languages akin 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 notably drawn to [Company I am intentionally blurring] dedication to creating accountable and dependable AI methods that prioritize buyer expertise and ease. My ardour for staying abreast of the newest AI analysis and my skill to judiciously apply novel strategies in manufacturing align with the corporate’s imaginative and prescient. I’m excited concerning the alternative to work with a cross-functional group 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 wonderful communication and presentation expertise, which have been demonstrated via my technical writing expertise at In the direction of Knowledge Science, the place I’ve written complete articles on machine studying and knowledge science, reaching a broad viewers of 50k+ month-to-month viewers.

Thanks for contemplating my software. 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 carry humanity and ease to banking via AI. I’m assured that my ardour for AI, my technical experience, and my skill to work collaboratively will make me a invaluable asset to your group.

Sincerely,

Piero Paialunga

They give the impression of being similar to I’d write them for a selected job description. That being stated, in 2025, you should watch out as a result of hiring managers do know that you’re utilizing AI to jot down 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 wish to say to use these instruments properly. Positive, you possibly can construct your “template” with them, however make sure to add your private contact, in any other case your cowl letter will probably 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 learn how to use LLM to transform your resume and job description into a selected cowl letter. These are the factors we touched:

1. Using AI in job looking. Within the first chapter we mentioned how job looking is now utterly revolutionized by AI. 
2. Massive Language Fashions concept. You will need to design the LLM APIs properly. We did that within the second paragraph
3. LLM API implementation. We used Python to implement the LLM APIs organically and effectively
4. The Net App. We used streamlit to construct a Net App API to show the ability of this method.
5. Limits of this method. I feel that AI generated cowl letters are certainly excellent. They’re on level, skilled and nicely crafted. Nonetheless, if everybody begins utilizing AI to construct cowl letters, all of them actually look the identical, or no less than all of them have the identical tone, which isn’t nice. One thing to consider. 

5. References and different good implementations

I really feel that’s simply honest to say numerous good those that have had this concept earlier than me and have made this public and obtainable for anybody. That is only some of them I discovered on-line.

Cover Letter Craft by Balaji Kesavan is a Streamlit app that implements a really comparable concept of crafting the duvet letter utilizing AI. What we do completely different from that app is that we extract the resume immediately from the phrase or PDF, whereas his app requires copy-pasteing. That being stated, I feel the man is extremely gifted and really artistic and I like to recommend giving a glance to his portoflio.

Randy Pettus has a similar idea as nicely. The distinction between his method and the one proposed on this tutorial is that he’s very particular within the info, asking questions like “present hiring supervisor” and the temperature of the mannequin. It’s very fascinating (and good) that you would be able to clearly see the best way he’s considering of Cowl Letters to information the AI to construct it the best way he likes them. Extremely advisable.

Juan Esteban Cepeda does an excellent job in his app as nicely. 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 critiques by customers. Nice job and nice hustle. 🙂

6. About me!

Thanks once more in your time. It means lots ❤

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 speak about AI, and Machine Studying in my weblog posts and on Linkedin. In case you appreciated the article and need to know extra about machine studying and observe my research you possibly can:

A. Observe me on Linkedin, the place I publish all my tales
B. Subscribe to my newsletter. It’s going to hold 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. Grow to be 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 concerning the latest expertise obtainable.
D. Need to work with me? Examine my charges and tasks on Upwork!

If you wish to ask me questions or begin a collaboration, go away a message right here or on Linkedin:

[email protected]