Inspiration
When have you ever looked forward to opening your inbox?
Us all being college students tracking assignments, job placements, and personal email all doing that simultaneously can make the whole process soo overwhelming and tedious. And imagine during your midterms (real story!) your dream company sends you an invite mail and that gets buried in the mail. This is genuinely a life hack that we believe can save us (and many others) lots of time!
What it does
It's in the name! autoMate, it comes into the picture and solves the problem of manual work by automating all your job-tracking processes. It scrapes your emails and filters the email based on the job and gets the company name, job role, and current stage of the interview process. This will help overcome the errors involved in manual data entry. This is all done using some clever heuristics and NLP Modelling which will classify your model whether you have been accepted/rejected at your application, when is the due date, and anything else that might be needed!
How we built it
We approached the problem by researching how we can parse the emails effectively. We found Gmail API to do the same. Then Machine Learning Fanciness: It uses in-house classification algorithms such as Naive Bayes and Count Vectors with 0.92 recall to converting email text into a model in which we determine the status of the application. We have also extensively used the Google Clouds Natural Language platform to basically extract relevant entities that can be used to extract Job Organizations, relevant job locations, and the deadlines to apply. We have also used large data sets accumulated from Kaggle to create a data dictionary in which we are searching for job roles and companies' lists which can be exhaustive items to identify whether it is part of a company or not. Sometimes you do not always need a fancy ML algorithm and algorithmic thinking can really help instead.
Challenges we ran into
- Our initial idea was to use a third-party client OAuth to fetch the emails. However, Google blocked 3rd party apps from accessing email data this year.
- Our next challenge was to distinguish between a 'job' email and a 'normal' email
- Maybe sometimes fussing over that minor coding detail does not make that big of a deal.
- Spread out the work, we all tend to work during the end, but if we all just do a constant amount of effort throughout the event, that seemingly daunting task becomes much easier.
Accomplishments that we're proud of
- First international in-person hackathon for many of us! Can you believe it? Built our very own ML algorithm with 0.92 recall, lots of room for improvement obviously!
- A lot of us were very new to the front end and made a decent one! We think we actually made a tool that maybe we can use! It is able to scrape emails and find the relevant job-related data like the company name, the deadline for the next round like online assessment, interview, and if the final result of the job application was rejected or successful.
What we learned
Time management is key, sometimes the most obvious/trivial solution is the best one. A lot about modeling challenges when building a practical solution and what parameters we can optimize on.
What's next for autoMate
Implement a complex classifier for each of the subproblems of the applicant tracking process (Categorizing emails, finding sentiments valence and nuanced measures, also adding additional features like URLs, tracking multiple applications from the same company, and so on! Also, we plan to create a task scheduler based on the next date of the interview/assessment. This would help the user not miss any deadlines.
Log in or sign up for Devpost to join the conversation.