Description as a Tweet:

DrBear is an assistant for all your investing needs. DrBear recognizes how volatile the market is, and how the market is defined by the people in it. DrBear is a chatbot that brings in the latest news, data, tweets about the company you want to see, using Sentiment Analysis.

Inspiration:

Investing is a task that needs a lot of background about any company. While many trading softwares do a cursory linear regression, such methods are useless at best and misleading at worst. A more informed user needs an assistant that assists and gives the necessary information. The market after all, is made of people, and what people like, sells. This is why we wanted an app that can assist users for investing decisions.

What it does:

DrBear is a chatbot assistant for finance news, discussions on forums, tweets and other such features. It uses Sentiment Analysis to predict the health of a company based on how people talk about it. It has a web-app which does real-time scraping and implementation of NLP techniques for the above. Our Github Readme has more detailed information about the project.

How we built it:

We have used Python for data-fetching and the Machine Learning models that we used, using various libraries such as tensorflow, pytorch, nltk, and employing use of Regex for data cleaning. Our web-app has a React frontend and a Flask backend, and is hosted on a Azure platform, employing Dialogflow from GCP libraries for our chatbot. https://imgur.com/a/L1IbATT has a basic flowchart of our methodology.

Technologies we used:

• HTML/CSS
• Javascript
• Node.js
• React
• Python
• Flask
• AI/Machine Learning
• Misc

Challenges we ran into:

Slow speed of scraping and of our model led us to do a lot of optimizations (further detailed on in our Github), the integration of several key components was also a challenge. Apart from this, figuring out how to run each API call was a time-intensive challenge, and we had to parse the data in a way that could be represented by our Web Application.

Accomplishments we're proud of:

We are proud of making a real-time scraper, and of our ML model which gives accurate results at high speed. We are also proud of our UI/UX.

What we've learned:

We got better at using a lot of these technologies and also on how to use Transformers efficiently. We also had to read about investing, as we did this as part of a proposed idea for a startup.

What's next:

Further improving speed, and implementing algo-trading solutions to add more depth to our bot. Hyperparameter tuning to improve our Model is also on our priority.

Built with:

We have used Python for data-fetching and the Machine Learning models that we used, using various libraries such as tensorflow, nltk, and employing use of Regex for data cleaning. Our web-app has a React frontend and a Flask backend, and is hosted on a Azure platform, employing Dialogflow from GCP libraries for our chatbot.

Prizes we're going for:

• Best Finance Hack
• Best Documentation
• Best Venture Pitch
• Best Web Hack
• Best Machine Learning Hack
• Best Use of Google Cloud

Mayank Goel
Kunwar Shaanjeet Singh Grover
Vishva Saravanan R
Tanishq Chaudhary

Table TBD