A fictional company called Doodle is looking to automate the recruting process of developers and wants an in-house solution to select the best candidates.
This repository presents a solution to Doodle's problem.
Our application retrieves data about possible candidates from various sources (see sources below) and classifies the candidates as selected for an interview or not. If a candidate is selected then an email is sent with a link to an interface with coding questions.
We used a combination of unsupervised learning and supervised learning to train the model.
First we implemented clustering to classify our users into two distinct categories, selected for interview and not selected. Then we proceeded to train a supervised model using logistic regression that would predict if a user was selected or not based on the labeled created through clustering.
The sources and API calls we used to get our developers were the following:
| Source | API Documentation |
|---|---|
| Stackoverflow | api.stackexchange.com/docs/users |
The retrieved data used to select candidates was stored in Mongodb Atlas.
The answers candidates gave in the coding questions from the interface were stored in an Sqlite database.
Mongodb was chosen given that the API calls return JSON files, which is an ideal format for this NoSQL database. It also allows for cloud storage with Atlas, and has great integration with Python through the py-mongo library.
Sqlite provides a simple SQL database to store our structured information, the interface users and their answers.
Our team used github projects to keep track of tasks and responsabilities. We divided our work the following way:
Legend:
✅ → responsible; ✔ → helped; ❌ → not involved.
Our current application has lots of potential improvements, these are a few of them:
- Select candidates from more data sources;
- Have a code coverage of 80% for unit tests;
- Connect both databases;
- Have a page in the interface for the recruiting manager to acess the answers;
- Be able to email the candidates directly from the interface;
- Landing page thanking candidates for the submission;
- More appealing timer.
Website:
Folder with everything related to the interface
preprocessing_tests:
Different tests and trials done while in the preprocessing phase, can be ignored
presentation:
The presentation was done in Markdown, all the code is provided there
utils:
Folder with the MongoDB setup
github.py:
Calls the github API and stores the data in MongoDB -> WORK IN PROGRESS
main.py:
Starts running the interface. It will run on the local machine
preprocessing.py:
Handles data preprocessing, creates and fits the model, at the end returns a dataframe with selected candidates
send_email.py:
Reads MongoDB collection selected_candidates and sends them an email with the interface link
stackoverflow.py:
Calls the stackoverflow API and stores the data in MongoDB
store_candidates.py:
Imports preprocessing.py and then stores the returned candidates in a new MongoDB collection called selected_candidates