Elie-lee
A research system that looks for companies worth owning for decades.
- Field
- Investing and artificial intelligence
- Covers
- 10,407 US-listed companies
- In depth
- 685 companies researched so far
- Reads
- Filings, earnings calls, insider trades and hiring data
- Built
- About 72,000 lines of code, with Claude Code
- Next
- Companies listed outside the US
Why I built it
Compounding over decades fascinates me. Eli is how I look for the few businesses that can do it.
You’ll get very rich if you thought of yourself as having a card with only 20 punches in a lifetime and every financial decision used up one punch.
The hard part is saying no to almost every company.
The six steps
ModelCodePerson
1Screen
Works by inversion: it looks for a reason not to buy, in the filings, the news and my notes on how companies fail. It can only say no.
Model15%of the companies researched stop here
2Classify
Works out what kind of business it is, because a software platform and a consumer brand can’t be judged by the same numbers.
CodeModel37%of those researched are kinds of business these steps don’t judge
3Analyse
Answers three questions: how does it make its money, what stops a rival taking it, and how long can that last?
Model
4Score
Grades the business against bars set before any company is seen, so a good story can’t move the grade.
Code
5Value
Picks the valuation method that fits how the business grows, with a margin of safety sized to its quality. A separate agent, which never sees the price, checks the reasoning and can only say no.
CodeModel6ways to value it (earnings power, owner earnings, discounted cash flow, asset value…)
6Decide
A final synthesis puts every finding in one memo. A fixed checklist confirms each step was passed, and the memo stays a draft until a person approves it.
CodePerson