Building in public - AI Job Application Agent

Firstly, I want to thank everyone who have been sending me personal messages encouraging me to write more. It means the world to me and I will keep the fintech series going with next post coming soon.

Today, however, I wanted to start a new series on #BuildingInPublic, where I'm going to attempt to build an AI agent to apply for jobs for me. Why this series? - Glad you asked. Reasons are simple for me

#1 It solves an immediate need for me. I'm looking for jobs in USA and spend countless hours going through the rigmarole of applications.

#2 There are many current tools in the market which do a bit of what I want but nothing seems to be solving for this properly.

#3 I hoping this tool can then me either used to help the community or better even be monetized.

What do I want to get out of this?

#1 I want to commit to something in public and actually build it.

#2 I want you my reader, to comment on something that you want for yourself. If I'm able to build it successfully, I would like to pass the knowledge on and help you too. So, if you are someone who are actively or passively looking for a job, spread this around by sharing, reposting, commenting (whatever is your cup of tea). My aim is to take as much feedback as possible.

Without boring you further, in today's article, I'm doing the step of building the agent which is, like any product, defining the requirement.

At the face of it, this seems simple - find jobs and apply right? Let's make it more nuanced. Note that this is my wishlist, and being the product manager I'm, I'll get into a prioritization once I research and gather more data points on what is possible, how much does it cost, what efforts will it take (you know RICE++...).

For now, I want the agent to act almost like an expert recruiter and that translates, in my mind as below

Step 1 - Understand the applicant

Problem here with current automated job appliers is that, this phase is completely superficial. Most tools ask for your current resume, job roles and location preferences. But in reality, this is much more detailed. For me this is as below;

What have they done so far?

While a human recruiter would conduct detailed interviews to get to this stage, for an AI to do this? It has to be iterative. I've gone through this phase with explicit prompting with ChatGPT and I want the AI agent to replicate the same. So what does this involve

(a) Without re-inventing the wheel - Get the user to upload their resume

(b) Scan the resume and start asking clarifying questions on each point mentioned in the resume. Attempt to capture all information required to put this in STAR format.

(c) Ask questions if the applicant did anything else in the role that would be considered important.

(d) Document point (c) and repeat.

(e) The end goal of above exercise is have an exhaustive list of what the applicant has done in their job which can be then refined for job openings. If put in a regular resume format, the information captured should fill a minimum of 1 page per year of experience the applicant has had.

(f) Last and most important step, connect to the users social media profiles. Example connect to LinkedIn, X so that connections can be identified in target companies automatically.

The other sections are self explanatory but I want the agent to intelligently prompt and probe the applicant and record everything there is to know about the applicant.

Bonus Features

(a) Suggest roles that are applicable to the geography - For example a business head in a fintech in India roughly translates to General Manager in USA

(b) Give the applicant a honest-to-god review on their aspirations. Ex. some one says that they are going for a VP of product role with 2 years of experience, tell them it ain't gonna happen.

By the end of Step 1, the agent should have a compendium of what has the applicant done so far and what do they want to do. We then move on to step 2.

Step 2 - Scout for Jobs

This step is a simply gathering all available jobs and prioritizing which ones to apply for;

Search for listings

The agent should ideally scour the internet, come back with listings and descriptions of jobs that meet the criteria. My wish list here, go above and beyond this means

(a) Obviously starting with Job Boards (LinkedIn, Indeed)

(b) Career Pages of Companies that have hired for similar roles in last X months.

(c) Recruiter Job Listings (Michael Page etc.)

(d) Social Media Posts of Hiring Managers, Recruiters etc.

Filter Listings

The agent should then filter the listing for criteria we made it aware in Step 1

Prioritize Listings

This where some magic needs to happen. Through iterative attempts I want the AI to learn which jobs to prioritize applications too. To start off with, some rules could be

(a) Apply for the recent postings

(b) Apply for the postings where the match rate between desired qualifications and the candidate is more than 80%

(c) Apply for postings where the applicant has referrals / friends options

Step 3 - Apply for Jobs

This involves multiple things and not just a simple application. Ideally I want the AI to be able to do below

(a) Obviously - Apply on the company website / LinkedIn automatically

(b) Customize resume and cover letter basis job requirements from the compendium it got in step 1. This customization should ideally include taking my input resume and creating an output resume in the same formatting I gave to it as a template including making sure the resume fits in 1 or 2 pages exactly (and not 1.37 pages). Creating a cover letter that is 5 paragraphs in length, not less or more etc.

(c) Find contacts it can message to from your LinkedIn/X Account (1st/2nd/3rd Connections) and shoot them a personalized message / email on why they should refer you for this job.

(d)Find HR/Recruitment/Talent Acquisition Manager profiles/emails and shoot them appropriate messages as well

Above are basic, next the wishlist begins

(a) Analyze the job opening - understand the company, product, role (in that order) to create a proof of concept project automatically and add it to your portfolio and add your portfolio link to the job application

(b) Message hiring manager, CXO in the firm with the link to the proof of concept project and explain why they should hire you.

Step 4 - Track, Track and Follow Up

There is a saying that what doesn't get measured doesn't get improved on. This is true both from model improvement perspective and job application perspective.

I want the model to create a comprehensive dump (ideally somewhere on google sheets and google drive) on

(a) Jobs its applied to

(b) Questions it answered on behalf of me in the application along with answers

(c) Artifacts it created for that job (Resume, Cover Letter, Proof of Concept Project, anything else)

(d) Status of responses from listing.

(e) Anything that needs my attention (did I miss a recruiter call!!)

Step 5 - The real real wishlist

#1 Catch Hallucinations - When I tried to do some of this manually using ChatGPT, I could really see hallucinations come into play. It invented things I did at my job which I never did and started adding it to my resume. I want every single point it puts in my resume to be cross checked against the compendium to make sure it is not lying for me in the resume.

#2 Tone/formality customizations - Start looking into the social media and website posts of companies it is applying for and adjusting the language / tone of resume/cover letter to suit the company more and echo its vision/mission/values

#3 A/B Testing - Start testing out templates, language, formats to see if a particular value yields better responses

Thats where I'm starting from folks. If you have read till here. Congrats! you have a great deal of patience. If you read this and felt there is something that you personally would like it to do as well, feel free to drop a comment or mail me at [email protected]