- 🟨 The Yellowbrick Road
- Pages
- 🤖 How The Yellowbrick Road AI Tool Works 🤖
Summary
We built an AI tool that reads and summarizes 100s of stock-related articles every single day to find the best stock pitches on the internet. We then pick the best ones and send AI-generated summaries of them in a free email every day.
It’s the best way to stay on top of all the great stock market content put out every day (and it saves us hours every day).
Sources
We are always adding more sources, but we currently read and summarize the following:
Seeking Alpha
Motley Fool
Twitter
Value Investors Club
Individual Blogs / Newsletters (both paid and free)
Podcasts
Investor Letters
Morningstar and Argus Analyst Reports (paid)
Any sources you want us to add next? Email me at [email protected] :).
How it Works
Our system is broken into 3 tools:
The Reader / Transcriber
The AI Summarizer
The Ranker
The Reader / Transcriber
Every day, our Reader tool goes out and reads all of the stock market articles from the above sources. The Reader is tailor-made for each source to accurately find the content posted for that day and then read the content from the article.
For written text on websites, our Reader just has to read the text for each article (pretty easy). For PDFs, our Reader has to download the PDF, filter out all of the disclaimers and non-relevant portions of the PDF, and then read the relevant PDF content (harder). For podcasts, our Transcriber first has to listen to the podcast and transcribe it (using OpenAI’s Whisper) and then read the transcript (even harder).
The AI Summarizer
After our Reader has read all of the content, the content gets passed to our AI Summarizer tool which summarizes the article. Once again, the Summarizer is tailor-made depending on the content.
How the AI Summarizer tool works:
The content from the Reader is passed into the AI Summarizer
The AI Summarizer checks which kind of content it is (pdf, text, html, etc) and chunks the content into smaller sizes that the AI can handle (we use LangChain for this step and the following step).
Once the content is chunked, it gets passed into one of our custom prompts based on the kind of content (we have different prompts for stock pitches, deep dives, interviews, etc) and fed to Open AI’s latest GPT model to summarize
The summary is then saved to be scored
The Ranker
After the content has been summarized, we use AI to rank the articles based on how compelling of an argument the AI believes the article made. All of that day’s summaries and ranks are stored in a single document for us to read through.
We then manually read through the summaries and ranks to decide which summaries should be included in that day’s emails. We have this manual step because we don’t always agree with our AI’s ranks or may choose to not include an article because the stock is not one we want to cover, is about a company that is already covered, etc.
—
If you have any questions, email me at [email protected]!
If you found this interesting, make sure to: