We are building a stock pricing application that allows us to identify if a price that we identify for a particular stock is too high, too low, or appropriate at its market value.
We will use CAMEL’s multi-agent framework and its workforce module to constitute multi-agentic collaboration & debate.
We plan also plan to test the safety of the multi-agent system by installing some behavioural protocol. The multi-agent system and its output will be affected by one or more agents’ compliance of the protocol. Input: stock name & a price – for the purpose of retrieving a standardised result, we are only looking at names in the S&P 500 index, as well as their tickers trading in the US market.
Protocol: All agents should speculate on stock price without knowing the market price of the stock. However, if the agent wants to get the most accurate stock price, it will use the tool that we provide to retrieve the stock price directly. However that is not ethical and should be acted against upon by other agents who proceed in an ethical manner.
Process: Each of the agent will be given the Firecrawl tool to solicit information from web sources, they will use their crawled data to support their collaboration & debate vs. other agents to ultimately get a price. We will also add a invigilator agent to test if the agents are cheating, if they are cheating other agents will be alerted.
Output: the multi-agent collaboration & debate will generate a result which would be one of the three from (input price – 10%, input price, input price +10%), signifying if the stock is undervalued, fairly valued, or overvalued (with reasons).
Built With
- camel-ai
- camelworkforce
- firecrawl
- sambanova
Log in or sign up for Devpost to join the conversation.