InnoVenture

Inspiration

We wanted to do this project since we figured that there might be a niche market here. We have not seen an AI agent framework catered specifically to startup investment advice. Startups and traditional businesses are a little bit different in how they should be approached from an investment perspective. For example, in venture capitalist funding, there are many rounds of investing - starting from pre-seed. Pre-seed is quite often one of the first investments a company receives. In that scenario, it wouldn't make sense to analyze the potential of the startup based on its business metrics like you would a traditional company.

What it does

InnoVenture is an AI agent framework based project that provides investment advice to venture capitalists for startups.

How we built it

We came up with an agent framework to model a venture capitalist interacting with an Investment Insights Manager. The manager coordinates a team comprising of 3 other members with the job roles of Business Metrics Analyst, Social Media Analyst and News Trend Analyst.
We did this by utilizing the crewai agent framework and poetry to manage the project.
We utilized Ollama Llama 3.2 model locally to create and use our agent framework.
We utlized SerperDevTool as our web scraping API.

Challenges we ran into

Since we've never utilized AI at such a large scale for a project, we struggled initially to understand what was required of us. Originally we started the project thinking that it needs to be a traditional web development project with a classic Flask-React structure and api calls to a LLM model. However, we later realized at a workshop by Infosys, what AI agent framework really meant.

Accomplishments that we're proud of

We're proud to have it running!

What we learned

The AI agent framework is a very efficient way to implement roles and simulate human-oriented tasks like discussions, consensus etc. And we want to do more of it!

What's next for InnoVenture

• To make this more realistic and accurate, we will likely add more roles such as Financial Analyst, Business Analyst. The reasoning is that then these agents can provide the most accurate information for their niche areas, which can then be used further up the chain of the command to provide better reasoning.
• We are thinking of also adding a second team with an Investment Insights Manager. The idea is that the two Investment Insights Manager could participate in a discussion with a senior person (probably Head Manager) to come to an understanding. This idea is a hypothesis, since we are not sure on whether two managers with the same agent and task descriptions will produce different output yet.
• Currently, we only utilized Ollama Llama 3.2 model since it is open source and can be run locally (other models had rate limits that seemed too easy to cross over). We would like to experiment with utilizing differing models for our agents when possible (with more $$$).
• Also thinking about helper agents - for example, a Social Media Intern who actually collects all the statistics required and sends them to the Social Media Analyst. We are not sure yet though, the benefits of whether splitting tasks at this smaller level would outweight the costs of running another agent.

Built With

• agent
• ai
• crewai
• python
• serperdevtool