Rounding up my AI coding work over the weekend

This is another power-up week, where I was able to dream up and execute some of the projects I've always wanted to do, but couldn't, because the coding was too hard or labour-intensive.

a) Built and grown my Second Brain

Obsidian is a humble note-taking app that's a sort of dumbed-down version of Notion. I noticed younger lawyers use it religiously, but I was never able to leverage it because I did not develop a habit of capturing information in my professional life.

So, after watching a few online videos and reading the Cheerful Egg blog, I decided to use Codex to populate my Obsidian vault with information on my projects, training materials, and content, and it was very satisfying to see the materials in graph view grow. I can now understand why it has gone viral - it's like growing a Mini-Me in a lab environment.

This experiment would have ended there, but I realised I could open this Second Brain to Codex or Claude Co-Work so it could pull information from all my previous work to create new content. 

b) Created a program to generate analyst reports

At this moment, the Early Retirement Masterclass already has a fairly mature skills markdown file that instructs AI to download historical financial information and produce an analyst report, along with target price estimates and dividend sustainability. Over the weeks, this MD file has been improved with new accounting metrics, such as Pietroski scoring, so the analyst reports for a stock are not small documents. They range from 35 pages to 70 pages long.

I want to evolve the skills markdown into a Python program that performs the exercise on all 80 stocks in the STI and SGX Next 50 lists. Getting all the reports took an entire evening and cost about $40 USD worth of tokens, and I've actually made it way cheaper by writing a program to download stock information into a database before I started running the Fundamental Researcher.

The final outcome is that, within a day, I have about 80 analyst reports, each 35-70 pages, on every mid-cap and large-cap stock in Singapore.

And AI analysts, having no need to pander to senior managers, are much stricter in their Buy recommendations - just look at the HOLD ratings below. The AI also has a value-investing bias, but that's because it makes decisions purely based on financial statements and business news results. 

I loaded all the reports to my Second Brain, and AI is just not impressed!

There are BUY recommendations in the end, but I think this is between my Early Retirement Masterclass community and me.

c) Built a legal case summary program

Given the sheer number of projects I could run on my machine, I felt confident enough to tackle some nagging issues I faced in the legal sector. 

And by now, this is an easy project.

I wrote a Python program that can take a legal case note from E-Litigation and summarise it into a 2-page study guide for students to just get the main facts on the case.

Then I got an ex-classmate to refer me to a case presided over by my Prof Goh Yihan, called Re: CK Tan Law Corp, and it not only provides a summary but also tells a law student how to use the case. 

Maybe I should go back to school for an LLM just to see how far this tool can take me. 

I think university students should not avoid reading the full case. In a Polytechnic environment, I can create a case book for my students to get the gist of the law, which is more than enough for a paralegal.

So, as you can tell, this has been a fairly action-packed weekend.

I've not thought about what this really means for the industry, given that a 52-year-old uncle can churn out 80 analyst reports and 5 legal summaries over a weekend and still celebrate Mother's Day with family. That can come in a future article.

But it's hard to describe the sheer power these co-working and coding tools bring to MBTI ENTJs like me. 

It's like commanding an army of elite developers who are at your beck and call 24/7. And if you are skilled at prompting and can think in terms of processes and step-by-step implementation, it can be amazingly addictive, as well as human beings often misinterpret the high-level commander's intent in the real working world.

Moving forward, I will likely channel more dividends into buying AI tokens, but I hope the sharper analytical tools I build will pay for them many times over as I achieve better investment results.