BunkIt

A ML curated excusee, so that you can bunk your next class smoothly😎

Comment

Bunkit

A ML curated excuse maker, so that you can bunk your next class smoothly😎

This project is built during first day back MLH hackathon.

What on earth is this?

We did not plan to start out planning to make this, but we wanted to align a bit with a theme as an afterthought, hence we thought, there must be some who still won't be feeling like attending classes, so why not help them!! And thus the idea of bunkit was born. It may seem deceivingly simple on top but under the mood, there's a lot going on.

Tech Stack

  • Flutter(app)
  • HTML+CSS(web page)
  • Python + Flask web app
  • gpt2 for text generation
  • Heroku for static excuses
  • GCP compute engine for gpt2 text generation

Installation

To build the flutter app, make sure you have android sdk and flutter sdk already set up, then.

  • Clone this repository git clone https://github.com/nayakastha/spikeORtom
  • cd app
  • 'flutter build apk -release'

just in case you are unable to do that here's the apk file

To build the web app

  • Open the root in shell and run cd web-gpt2
  • Install the required dependencies pip requirements.txt (Warning, large download size)
  • To start dev server, run python app.py

Functionalities

There are 2 types of excuses this generates

  • A static excuse, which is basically a random combination of some excuses we have hardcoded, structured to give a syntactically correct sentence. You can get this by going to localhost/getx or from the live server here
  • The other (fancier!!!) excuse is generated when we pass some random keywords from the above excuse to a gpt2 model. It gives out coherent sentences, with some absurdity of course(this baby is still learning how to make excuses). Each excuse takes 8-10s to generate so kindly bear with us for now. You can check it out on your localhost localhost/excuse/gen or from the live server here

Challenges Faced

  • One major problem I faced is not getting GCP credits because we are from APAC region, hence not eligible.
  • Running a gpt2 model on a live server is very intensive an affair, with CPU usages sometimes going over 140% of stipulated firepower. We're using a e2-medium(2 vCPU, 4Gb memory, Intel Haswell), so it is burning through the little GCP credits we had really quickly, a solution to this would be welcome.
  • This was the first instance we have worked on a web project, let alone a ml and gcp project, those were far fetched until now. The server is not yet fully optimized, and crashes at randomly.

Built With

Share this project:

Updates