Interest Rate Predictions Exposed: Analyzing YouTube Financial Forecasts or Is this channel "Full of Beans"

#devchallenge #agentaichallenge #ai #machinelearning

This is a submission for the Agent.ai Challenge: Assembly of Agents (See Details)

What I Built

I have always had an interest in Financial commentators. Specifically, how accurate are they? Before YouTube those old predictions were unavailable. With these tools we can now retrieve the transcripts of old comments and with the LLM's we can easily bang those predictions against what actually happened. That is my vision for this agent, test pundits predictions against reality.

Because I came from a coding background I was looking for a method to structure my agent. That structure came from calling sub-agents. Thus it was natural to submit this agent to the Assembly of Agents contest.

Demo

Here is my agent: Link

Here is a video demonstrating its operation, output, and program structure. Link

Agent.ai Experience

I would not have found agent.ai without DEV. The announcement popped up in my daily.dev feed and I was hooked! Nothing like a good hackathon to get my keystrokes flying.

Having made some contributions to Fabric, an agent.ai competitor, I was somewhat familiar with the agent concept from a agent developer viewpoint.

Overall, I really like the agent.ai platform and functionality. Coming from a coding background, several concepts were redesigned by agent.ai for more LLM compatibility.

I found that by using sub-agents, I was able to introduce some structure into my code. I am still puzzling over the variables and looping as envisioned by agent.ai.

I wish for "system" calls, like time and date and for me to document my code as this agent took about 40 lines not counting the sub-agents I called, a commenting action would be helpful.

I am an individual contributor under the DEV username duckpaddle.