Constructing a Resume Overview Agent System with CrewAI

Crafting the right resume is a tedious process – whether or not you’re a contemporary graduate getting into the job market, or a seasoned skilled aiming for that subsequent massive profession transfer. However what should you may have a private resume reviewer at your fingertips – with out spending any cash on LinkedIn Premium or hiring an expensive skilled? Enter the world of AI-powered options for resume optimisation! With the facility of Massive Language Fashions (LLMs) and modern libraries like CrewAI, now you can construct your very personal resume evaluate agent. On this weblog, I’ll present you methods to construct an agentic system with CrewAI for reviewing and optimizing resumes.

Construction of the Resume Reviewer Agentic System

Earlier than moving into the coding half, it is very important perceive methods to construction an AI system for resume optimisation. One of the best ways to do that could be by noting down the duties that we want the agent to carry out. So, I’m assuming you’ve got already made a resume (else you possibly can obtain one resume from right here). Now, we’d need our resume evaluate agentic system to carry out 3 duties:

1. Learn by way of and supply suggestions on the resume.
2. Enhance or re-write the resume based mostly on the suggestions.
3. Recommend applicable jobs based mostly on the improved resume and specified location.

Now that now we have our necessities clearly written, we are able to determine on the variety of brokers and their duties. Ideally, it is suggested we go together with one process per agent to keep away from overburdening any agent. This interprets to us constructing 3 brokers for our CrewAI-based resume evaluate agent system:

1. The primary agent will present suggestions on the resume.
2. The second agent will enhance the resume based mostly on the suggestions.
   
3. And, the third agent would recommend applicable jobs based mostly on the improved resume and specified location.
   

Additionally Learn: Constructing a RAG-Based mostly Analysis Assistant Utilizing o3-mini and CrewAI

Now that you simply perceive how we are able to use AI for resume optimisation, let’s soar to the Python code and construct our resume reviewer agent.

Python Code to Construct Resume Reviewer Agentic System with CrewAI

Listed here are the step-by-step directions to construct a resume evaluate agent with CrewAI. To higher observe these steps, I recommend you watch this hands-on video parallelly.

So let’s start!

Step 1: Set up and Import Related Libraries

We’ll start with putting in the PyMuPDF, python-docx and most significantly CrewAI.

• The PyMuPDF library is used for studying PDF paperwork.
• The python-docx library is used for creating, studying, and modifying Microsoft Phrase (.docx) paperwork.
• Additionally, guarantee CrewAI is put in in your system. It is without doubt one of the hottest agentic frameworks to construct multi-agent methods.

#!pip set up PyMuPDF
#!pip set up python-docx
#!pip set up crewai crewai-tools

Step 2: Loading the Resume

The following step is to import the fitz and docx modules. Right here ‘fitz’ is the identify you utilize to import PyMuPDF, and ‘docx’ is to import the python-docx library.

import fitz  # PyMuPDF for PDF processing
import docx  # python-docx for DOCX processing

We’ll now outline the next three capabilities to allow our agentic system to extract textual content from resumes saved in varied codecs.

• The primary operate – “extract_text_from_pdf” will extract contents from the resume in PDF format.
• The second operate – “extract_text_from_docx” will extract contents from the resume in docx format.
• The third operate – “extract_text_from_resume” wraps the primary two capabilities and makes use of the respective operate based mostly on whether or not the doc is a PDF or docx.

Let’s run this.

def extract_text_from_pdf(file_path):
    """Extracts textual content from a PDF file utilizing PyMuPDF."""
    doc = fitz.open(file_path)
    textual content = ""
    for web page in doc:
        textual content += web page.get_text()
    return textual content

def extract_text_from_docx(file_path):
    """Extracts textual content from a DOCX file utilizing python-docx."""
    doc = docx.Doc(file_path)
    fullText = []
    for para in doc.paragraphs:
        fullText.append(para.textual content)
    return "n".be part of(fullText)

def extract_text_from_resume(file_path):
    """Determines file kind and extracts textual content."""
    if file_path.endswith(".pdf"):
        return extract_text_from_pdf(file_path)
    elif file_path.endswith(".docx"):
        return extract_text_from_docx(file_path)
    else:
        return "Unsupported file format."

Subsequent, let’s check this operate with 2 resumes. The primary one might be a resume in PDF format of an individual named Bruce Wayne (not the superhero).

res1 = extract_text_from_resume('/Customers/admin/Desktop/YT Lengthy/Bruce Wayne.pdf')
print(res1)

Now, we view the second resume.

res2 = extract_text_from_resume('/Customers/admin/Desktop/YT Lengthy/Anita Sanjok.docx')
print(res2)

Step 3: Making ready the Brokers and Duties

This step is the place CrewAI enters the scene. We’ll now import crewai and begin constructing the agentic system. For this, we’ll want 3 parts from CrewAI, namely- Agent, Process and Crew.

• Agent represents an AI assistant with a particular position and objective.
• Process defines an goal that the agent wants to perform.
• And at last, Crew is used to bundle a number of brokers and their respective duties collectively to work and obtain the set goal.

import os
from crewai import Agent, Process, Crew

Subsequent, we have to add the OpenAI API key, which I’ve saved in a separate file. So, we have to learn the API key from the file and set it as an setting variable. On this case we’re utilizing GPT-4o-mini because the LLM.

with open('/Customers/apoorv/Desktop/AV/Code/GAI/keys/openai.txt', 'r') as file:
    openai_key = file.learn()
os.environ['OPENAI_API_KEY'] = openai_key
os.environ["OPENAI_MODEL_NAME"] = 'gpt-4o-mini'

Subsequent, we outline the brokers and their respective duties.

1. Resume Suggestions Agent

Our first agent would be the “resume_feedback” agent. The agent class has some parameters which assist us outline the target of the agent. Lets have a look at them:

• The position parameter defines the agent’s operate and experience throughout the crew. Right here the agent’s position is that of a “Skilled Resume Advisor”.
• The objective is the only goal that guides brokers’ decision-making.
• Verbose = True allows detailed execution logs for debugging.
• The backstory offers an in depth description of the traits of the agent. It’s just like defining the context and character of the agent.

Let’s run this.

# Agent 1: Resume Strategist
resume_feedback = Agent(
    position="Skilled Resume Advisor",
    objective="Give suggestions on the resume to make it stand out within the job market.",
    verbose=True,
    backstory="With a strategic thoughts and one eye on element, you excel at offering suggestions on resumes to focus on probably the most related expertise and experiences."
    )

Now let’s outline the duty for the resume_feedback agent. The duty is nothing however the particular task that must be accomplished by the agent to which this process is assigned.

• We’ll add an in depth description which needs to be a transparent, and concise assertion of what the duty entails. It’s essential to observe that the {resume} written in curly braces is a placeholder that might be dynamically changed with an precise resume enter when the duty runs. You possibly can pause studying right here, undergo the outline and make adjustments as per your necessities.
• The anticipated output parameter helps us decide the format of the output. You may get the output in varied codecs corresponding to  json, markdown or bullet factors
• The agent parameter highlights the identify of the agent to which this process is assigned.

# Process for Resume Strategist Agent: Align Resume with Job Necessities
resume_feedback_task = Process(
    description=(
        """Give suggestions on the resume to make it stand out for recruiters. 
        Overview each part, inlcuding the abstract, work expertise, expertise, and training. Recommend so as to add related sections if they're lacking.  
        Additionally give an general rating to the resume out of 10.  That is the resume: {resume}"""
    ),
    expected_output="The general rating of the resume adopted by the suggestions in bullet factors.",
    agent=resume_feedback
)

2. Resume Advisor Agent

We’ll do the identical for the subsequent agent, that’s, the resume_advisor agent which writes a resume incorporating the suggestions from the resume_feedback agent and defines its process within the resume_advisor_task. Be happy to undergo it.

# Agent 2: Resume Strategist
resume_advisor = Agent(
    position="Skilled Resume Author",
    objective="Based mostly on the suggestions recieved from Resume Advisor, make adjustments to the resume to make it stand out within the job market.",
    verbose=True,
    backstory="With a strategic thoughts and one eye on element, you excel at refining resumes based mostly on the suggestions to focus on probably the most related expertise and experiences."
)
# Process for Resume Strategist Agent: Align Resume with Job Necessities
resume_advisor_task = Process(
    description=(
        """Rewrite the resume based mostly on the suggestions to make it stand out for recruiters. You possibly can modify and improve the resume however do not make up information. 
        Overview and replace each part, together with the abstract, work expertise, expertise, and training to higher mirror the candidates skills. That is the resume: {resume}"""
    ),
    expected_output= "Resume in markdown format that successfully highlights the candidate's {qualifications} and experiences",
    # output_file="improved_resume.md",
    context=[resume_feedback_task],
    agent=resume_advisor
)

3. Job Researcher Agent

Now, the third agent ought to be capable to recommend jobs based mostly on the {qualifications} and the popular job location of the candidate. For this, we’ll grant a software to our subsequent agent that allows it to go looking the web for job postings at a location.

We’ll use CrewAI’s SerperDevTool which integrates with Serper.dev for real-time net search performance.

from crewai_tools import SerperDevTool

Now we import the API key from Serper. You possibly can generate your free Serper API key from https://serper.dev/api-key.

with open('/Customers/apoorv/Desktop/AV/Code/GAI/keys/serper.txt', 'r') as file:
    serper_key = file.learn()

os.environ["SERPER_API_KEY"] = serper_key

search_tool = SerperDevTool()

Now we outline our ultimate agent which is the job_researcher agent. This agent will seek for jobs on the location based mostly on the resume improved by the resume_advisor agent. The construction of the agent is just like the above brokers with the one distinction being the addition of a brand new parameter, which is instruments. Instruments allow you to assign the required instruments to an agent which helps within the completion of the duty. Additionally, within the process now we have added {location} in curly braces for dynamically altering it.

# Agent 3: Researcher
job_researcher = Agent(
    position = "Senior Recruitment Guide",
    objective = "Discover the 5 most related, not too long ago posted jobs based mostly on the improved resume recieved from resume advisor and the situation choice",
    instruments = [search_tool],
    verbose = True,
    backstory = """As a senior recruitment guide your prowess to find probably the most related jobs based mostly on the resume and placement choice is unmatched. 
    You possibly can scan the resume effectively, determine probably the most appropriate job roles and seek for the most effective suited not too long ago posted open job positions on the preffered location."""
    )
research_task = Process(
    description = """Discover the 5 most related latest job postings based mostly on the resume recieved from resume advisor and placement choice. That is the popular location: {location} . 
    Use the instruments to assemble related content material and shortlist the 5 most related, latest, job openings. Additionally present the hyperlinks to the job postings.""",
    expected_output=(
        "A bullet level record of the 5 job openings, with the suitable hyperlinks and detailed description about every job, in markdown format" 
    ),
#    output_file="relevant_jobs.md",
    agent=job_researcher
)

Step 4: Creating the Crew and Reviewing the Output

Now we attain the ultimate step the place we bundle the brokers and the duties so as throughout the Crew performance of crewAI. It has 3 parameters:

• The brokers parameter lists the brokers within the order through which they are going to be referred to as and executed.
• The duties argument lists the duties outlined for every agent within the order of execution.
• And at last, Verbose=True allows you to see the detailed output so you possibly can see what the brokers are doing.

Let’s run this.

crew = Crew(
    brokers=[resume_feedback, resume_advisor, job_researcher],
    duties=[resume_feedback_task, resume_advisor_task, research_task],
    verbose=True
)

And we lastly launched our agentic system with the kickoff performance.

• crew.kickoff(inputs={…}): Begins the CrewAI execution, and takes in 2 inputs, resume and placement.
• “resume”: res2: Helps you specify the resume for which the agentix system must work. In our case, we’re engaged on Anita’s resume.
• “location”: ‘New Delhi’: Specifies the place Anita is searching for a job. And this might be New Delhi in our case.

result1 = crew.kickoff(inputs={"resume": res2, "location": 'New Delhi'})

Looks as if our crew capabilities completely. Let’s print every agent’s output individually.

from IPython.show import Markdown, show

First, we’ll print resume_feedback agent’s output.

markdown_content = resume_feedback_task.output.uncooked.strip("```markdown").strip("```").strip()
# Show the Markdown content material
show(Markdown(markdown_content))

Then we print the resume_advisor agent’s output.

markdown_content = resume_advisor_task.output.uncooked.strip("```markdown").strip("```").strip()
# Show the Markdown content material
show(Markdown(markdown_content))

And at last, we print the research_task agent’s output.

markdown_content = research_task.output.uncooked.strip("```markdown").strip("```").strip()
# Show the Markdown content material
show(Markdown(markdown_content))

And there you’ve got it. Your totally useful resume evaluate agentic system with CrewAI.

Additionally Learn: Automating E-mail Responses Utilizing CrewAI

Turning your CrewAI Resume Reviewer Agentic System right into a Net App

One attention-grabbing factor we are able to do with our AI-driven resume optimization system is constructing a web-application for it.

Right here’s the interface of the app I constructed:

Now we have wrapped our code for resume evaluate agent with CrewAI in Gradio and hosted the ultimate product on Hugging Face Areas which has a liberal free tier.

Right here’s the way it works:

1. First, drop your resume in a PDF or Phrase doc. The identify of our candidate right here is Bruce Wayne.
   
2. Subsequent, choose the popular job location for Bruce who’s our candidate. Let’s say San Francisco
   
3. And hit Submit.
   

Since that is utilizing LLMs on the backend, GPT-4o-mini in our case, it would take a while to judge your resume, give a rating, and suggestions.

And as you possibly can see, Bruce acquired a rating of seven/10. And he has obtained in-depth section-wise suggestions from the CrewAI agentic system. Together with that he additionally acquired the revised resume and even strategies for jobs in San Francisco. How wonderful is that this!

If you wish to replicate this CrewAI agentic system within the type of this web-application, you possibly can merely clone this repository or duplicate this house and alter the API key. Right here’s how that’s achieved:

1. First, click on on Settings after which click on on Clone repository within the dropdown.
   
2. Then scroll all the way down to the variable and secrets and techniques setting.
   
3. You’ll already discover the variables we created for the OpenAI key and Serper_Key. Merely click on on Substitute and add the API Key within the Worth. Then click on Save.
   

Additionally, you will get the code to construct this app from the CrewAI agentic system together with the Gradio code within the app.py file in information part of Hugging Face areas.

The clicking on app.py

Doable Enhancements on the Resume Reviewer Agent

Like every other agentic system, there may be a whole lot of scope for increasing the resume evaluate agent we’ve simply constructed with CrewAI. Listed here are a number of tricks to tinker round with:

• You possibly can construct a fourth agent that customizes the resume, based mostly on the job you choose.
• You possibly can construct one other agent in the long run to generate a canopy letter personalized for the job you need to apply.
• It’s also possible to construct one other agent that can allow you to put together for the job you choose.

Additionally Learn: Learn how to Construct an AI Pair Programmer with CrewAI?

Conclusion

Constructing a resume reviewer agentic system is a game-changer for job seekers searching for resume optimisation with AI-driven insights. This method effectively analyzes, refines, and suggests jobs based mostly on customized resume suggestions—all at a fraction of the price of premium resume optimizing companies. With CrewAI’s agentic capabilities, we’ve streamlined resume optimisation into an automatic but extremely efficient course of. However that is just the start! CrewAI is engaged on automating all types of duties and bringing much-needed effectivity to all industries. It’s certainly thrilling occasions forward!

Regularly Requested Questions

Apoorv is a seasoned AI and Information Science chief with over 14 years of expertise, together with greater than a decade targeted on Information Science, Machine Studying, and Deep Studying. Because the Head of Coaching at Analytics Vidhya, he has spearheaded the event of industry-leading AI applications, together with programs on Generative AI, LLM Brokers, MLOps, and Superior Machine Studying, shaping the talents of hundreds of pros.