This open source repository serves as a companion to my interview bot, which answers arbitrary interview questions by selecting the most relevant response from a set of pre-written answers. While I have not freely licensed the demo site and accompanying notebooks, anything you find in this repository is free to use, subject to the MIT License. You can read more about my reasoning for this here.
If you have just downloaded this repository on your desktop or laptop, use the bash shell to run source setup.sh. This will create and activate a python virtual environment and install the necessary dependencies for you to run and test the code on your machine. Whenever you update a file in the data directory (for example, to use a different language model or change the chatbot's answers) you will need to run utils/refresh_cloud_function_data.sh.
Once you have run the setup script, start up the development server by navigating to the cloud-function directory and running functions-framework --debug --target="route_requests". Then, you can test the function using your browser to visit http://localhost:8080 for the default response, or http://localhost:8080?q=what+is+your+dream+job to get the answer to the question "what is your dream job?"
If you wish to deploy the chatbot as a Google Cloud Function, you can follow this tutorial.
There are three main directories in this repository: cloud-function, which is the back-end for the interview bot demo site, data, which is the data used to create the interview bot, and notebooks, which contains open-source versions of my notebooks. There is also a utils directory which contains some useful scripts for managing the repository.
As the name suggests, this directory is meant to be deployed as a serverless cloud function. Once deployed to the cloud, the software in this directory will accept users' questions and respond with appropriate answers. matcher.py contains the code for matching users' questions to pre-written answers, while main.py provides the main http interface. The cloud-function/data contains the data needed in production for the cloud function to work and is copied over from the main data directory.
In this directory, questions.tsv contains the entire corpus of interview questions I used for creating the chatbot, while answers.jon contains a smaller mapping of "golden" interview questions to pre-written answers. The answers are coppied to cloud-function/data. From there, cloud-function/matcher.py finds which question in answers.json is most similar to a user-submitted question and returns the answer corresponding to the most similar example question.