*2025 07 28 Daily Log

2PM

Day 1 of souping up my report generation scripts so that it can do all my work and I go back to complaining about Claude Code.

I have a pipeline for generating a few reports, but it’s quite brittle. I have to download exact reports, specify url, it is quite rigid in function structure so there’s one custom script per report type I’m generating.

It can be so much better. And there are ways to integrate claude code here to do some parts of the work.

Current goal: write a simple CLI program + server combo that can„

  • Pull latest KPIs from a long list of companies
  • Generate a simple comps sheet.
  • Do deep screening - go through a list of companies and answer yes/no on certain flags being present (e.g. ‘hidden gem segment)
  • Produce initial set of documents I use when looking at companies.
  • Summarize recent news flow.

Let’s get started.

2.15PM

First, a list of major back end utilities / services I need.

  • Search companies database
  • List companies in my watchlist
  • List filings - currently these are pre-categorized by keyword.
  • List transcripts
  • Select appropriate filings / transcripts - this is where I’ll plug in Claude for some ‘automated intelligence’
  • Render or download filings and save to local drive.
  • Send to gemini with an appropriate prompt template, depending on the routine

Most of these I already have from my previous scripts - just have to port them over.

2.30PM

Porting over functions taking longer than expected. Yes, it’s pretty easy to just copy paste but my old code base reads like a teenager wrote it in 1970s on a napkin. I want to take this oppt to organize this better.

Current file tree: imgDiary 2025-07-28.png

ontinuing to set up the main functions.

Current status

  • Filing list, including real time scrapes where necessary: done
  • Filings downloader: done
  • Database operations: done
  • Gemini service: done

What’s left?

  • Report generator
  • PDF handler? Esp python scripts.
  • Using these primitives to create a routine. I wonder if there’s a way to abstract all of these functions to make it much easier to create custom routines on the fly. Look this up

Gotta take kiddo to the library.

5PM

Back and have a bit of time before dinner. Let’s pick back up.

OK - just spent some time wrapping up my functions in classes in order to make it easier but looking at what I’ve done it seems entirely useless?

And - out of quick time.

8PM - post dinner

Well - that was a lot of nothing lol.

Testing out functions - works.

OK now - testing out the CLI UI - build it out and see what I need.

Wait - while this is cool, I might be doing things the wrong way? I’m building stuff out that I might not need. Have to REALLY CLEARLY DEFINE what is MVP here. Otherwise I’m just gonna end up building some cool looking CLI that I ultimately can replace with fasterdd.

What am I building here? Why am I building this? I want to automate report generation. The type of report generation where I don’t have to manually select filings - it will just auto do it for me.

Stretch goal is to have all the primitives built so that I can use it to build quick databases without manually going through stuff one by one.

10PM

Thinking around this a bit more - I think I can chunk what I’m trying to do into three buckets:

  • Select a list of companies to go through.
    • Select from watchlist
    • Select from universe (all companies under )
  • Run a routine - report gen / data extraction
  • Do stuff with data

Company selector

  • Watchlist - done.
  • Screener - done.
  • ID / Name - need a comma - and need to test out the picker.

Once I have this, I will need to figure out how to queue and run this - and how to resume if something fails. also maybe limit the total batch number - or warn at least - to prevent on accidental cost overruns. maybe batch process things as well?