Democratization of markets has increased exponentially over the last few years. Legions of young people are diving into the world of finance; and I wanted to make a platform that would make it easy for people like me to see all the information I need, and pick stocks based on things like fundamentals, technical or sustainability

What it does

Minerva learns for you: it aggregates realtime stock data along with message forums and tweets about the performance* of a particular business. It uses powerful Machine Learning models to summarize and classify these articles to determine if the company's performance merit’s a buy, hold or sell strategy when it comes to stocks. *Performance includes things like fundamentals, technical, environmental impact etc.

How we built it

Appsmith, Python, Scikit-learn, Pandas, TensorFlow, MongoDB, Javascript, Yahoo Finance API, Beautiful Soup, and Twitter API

Challenges we ran into

Featurization of text data. Started with bag-of-words model, but accuracy was too low. Moved to tf-idf but had erros with the featurization vector. Fixed this by using the same featurization vector for both the training and test data. Connecting realtime and uploading to MongoDB to make a seamless web interface. One man team makes it difficult to expand on all ideas.

Accomplishments that we're proud of

Getting things working. Backend fully developed. Machine Learning paramters tuned for optimal performance.

What we learned

Using appsmith to easily deploy a project like this. API calls are limited. Using databases and hyperparameter tuning in ML models

What's next for Minerva

Adding a trading function, more seamless real time data and better visualization and creating an Exchange Traded Fund made up of companies with strong fundamental/technical/sustaintability metrics.

Built With

Share this project: