Rizk | Devpost

Inspiration

There is no way for a retail investor to express their views in the form of investing. Retail investors are presented with either buying an index that someone else believes in or doing a lot of research to find stocks aligned with their views. We simplify the second option by generating a portfolio that aligns with the views of our investors.

What it does

Given any theme with input taken as a natural language string, finds stocks about the theme and builds a portfolio based on the expected return and tolerated risk factor.

How we built it

We used the attention mechanism to build embeddings for both financial news and stocks. We then computed the correlation using these embeddings. The encoder for news articles is also used for themes and we output the matching stocks. The risk is minimized as the quadratic form of the portfolio subject to the expected return. The front end was developed using Streamlit.

Challenges we ran into

We had some trouble pinning down a version of this idea that we could finalize. We started building a copilot for financial advisors but decided that we shouldn't limit the model to financial advisors and opened it to the public. In the technical sense, aggregating data from various sources was difficult with proprietary data being hard to get.

Accomplishments that we're proud of

Getting a demo ready despite our last-minute pivot: we managed to build encodings of asset holdings and use it to provide customized investment portfolios to common folks.

What we learned

We have to figure out what products might be more or less interesting to users. We got better at communicating and understanding each other's ideas. Moreover, we also learned that transformers are absurdly effective.