Kuber - your own personal financial advisor.

Empowering Your Financial Future: Personalized Advice for Smarter Decisions. Our advanced AI analyses your unique financial situation to offer not just the answers you seek, but what you need.

Comment

Inspiration

We took the inspiration from our own personal needs. Often everyone is in a situation to seek financial advice from someone who is not only an expert but also someone who can suggest what is best for themselves based on their own financial situations and goals. However, based on our findings and observation, we saw this as gap as today we can get financial advice but not something which is apt for me and even if it is there, it lacks in veracity and variety.

What it does

Kuber - the personalised financial advisor is a personal companion to anyone or more like a talking partner with whom anyone can talk to seek advice on their queries related to Finances. For example, what would be the best investment strategy for a user in short term? Should a user invest in one particular stock or not for a given time? For any kind of tax implications, what should be considered and what could be the approach. Kuber is an agent that registers user's profile to understand the financial situation, user's financial traits and characteristics and then answers any query from the user just like a financial guardian and not just like an advisor.

How we built it

We built is using a combination of many things. We used llamaindex to build a backend agent (essentially a RAG) that does the portfolio analysis including the stock analysis and then executed it as a service using GCP's Cloud Run. We exposed the service of this backend agent using an OpenAPI interface which is then being consumed by an vertex ai agent. The Idea is to expand this idea in future for various financial domains wherein we can have multiple fine tuned, RAGs representing multiple financial domains such as Tax, Stocks, Startups, General news etc. being invoked by a super agent such as in this case the one which is based on vertex ai agent. We also integrated the 3PPMoreover, we also captured user's insights as part of user registration process which can be updated on regular basis. For instance refer the following:

How would you describe your current knowledge of financial markets?: Intermediate - I understand the basics of investing and have some experience. You have received an unexpected sum of $10,000. How would you most likely use it?: Research different investment options and invest the entire amount. How would you describe your risk tolerance?: Moderate - I am willing to take on some risk for potentially higher returns. Which statement resonates most with your view on financial planning?: I prefer to maintain a balance between enjoying life now and securing my future. How often do you review your budget and track your expenses?: Rarely or never. When making significant purchases, how do you typically approach the process?: Impulsively, based on immediate desire. What motivates you to save and invest?: Fear of financial instability. Imagine you're planning a vacation. Which aspect is most important to you?: Luxury and comfort, regardless of the cost. How comfortable are you with delegating financial decisions to a professional?: Not comfortable, I prefer to manage everything myself. What are your preferred sources of information for learning about personal finance?: Popular online blogs and articles.

There's an agent that analyses user's financial situation and investment goals based on the above and grounds the results against general best practices in financial world. For that we used special sources such as books and blogs from some of the most reputed financial authors in the world.

Challenges we ran into

There were many challenges. The most worrisome were the following:

  • Vertex ai agents are not as per the expectation. Often they stopped working altogether, or won't yield anything and not even supported by a good enough error reporting.
  • The instructions and tool selection is not as apt as it looks in vertex ai agents.
  • While working on backend agents, we couldn't find a great support for gemini models in llama-index for example, while creating reasoning loops llamaindex has a great support for Openai models but not for Gemini.

Accomplishments that we're proud of

For a limited set of information, we were able to successfully test our idea of personalised financial advisor.

What we learned

Many things especially related to vertex ai agents, vertex ai, financial domain and much more.

What's next for Kuber - your own personal financial advisor.

This has a lot of potential and we are planning to extend it for more financial domains, better evaluation and to come out with results which can be consumed in other offerings such as websites, apps etc.

Built With

Share this project:

Updates