YWR GP: YDX (your AI Sell-Side)

I couldn’t come up with just one stock idea for you today.

So I came up with 393. I hope that’s OK.

Oh, and I built a website for you with all the write ups.

But before I get to YDX let me explain the background, and how we got here.

In the beginning we created the global factor model dashboard on Retool so we could scan global estimate revisions for 10,000 stocks each month. It helps us find trends nobody is talking about.

This dashboard and the data became the basis for our monthly factor model trends analysis.

YWR Factor Model Dashboard

Then to help improve our entry points we created the estimate/price scatter on Tableau to find stocks where estimates were rising, but price hadn’t moved.

Then we created the QARV dashboard on Retool to scan for high quality stocks trading at relatively attractive valuations.

Then we added the automated AI Reports button to the Factor Model and QARV dashboards so if we saw something interesting with 1 click we could get a 5 page deep dive. These AI reports included an earnings call section where we used AI to analyse how the company earnings calls were evolving over the last 3 quarters. Not ‘summarise the call’, rather ‘tell me the top concerns in the Q&A and if management sentiment is improving’ over 3 quarters.

Next we moved on from quantitative screening to see if we could use AI to build qualitative screening frameworks. This is the fascinating new possibility with AI.

For example, we loved the book ‘Blue Ocean Strategy’ and wondered, can we use AI to analyse and score 2,000 stocks for Blue Ocean shifts? This inspired us to create the YWR Blue Ocean Dashboard.

Then we read ‘Own the Rule Breakers’ by David Gardner, which we also loved, and used AI to screen qualitatively for his top 6 attributes. We created the YWR Rule Breaker Dashboard so we could screen quantitatively for ‘Top Doggedness’ or ‘Valuation Skepticism’. Remember David loved when investors didn’t want to buy high growth, founder led companies in growing industries because they were ‘expensive’. So we used AI to screen for this skepticism. We could also use AI to screen for ‘corporate culture’ if we wanted. This is all new. Things we could never do with quantitive screening.

Then came ClawdBot and we wanted one too. We wanted our own custom investment AI connected to as many financial tools, API keys and capabilities as possible. So we created ‘Stevie’, our own AI investment analyst on Telegram with access to the YWR data, financial data, options data, its own internal read/write database, Coinbase Wallet, and its own Google Workspace account.

We also gave Stevie tasks to do overnight, like listen to earnings conference calls, so he had interesting snippets for us in the morning.

So what’s next? What’s the new, new?

YDX.

Let me explain.

What problem are we trying to solve?

Our challenge in asset management is always how to increase our perspective. We may have found a great idea, but is it the best idea? How does it compare to all the other stocks out there? How do we make sure we are not late to the trends?

That’s the challenge of the active manager.

Find good ideas + make sure they are better than the 1,999 other ideas out there. It’s 3D chess.

But how do you do that unless you cover every stock in the market?

Maybe AI can help. What if we can automate AI to analyse hundreds of stocks on its own?

What if we could get the AI to build financial models, listen to conference calls, create earnings forecasts and write an investment case on its own? And if this was possible, could we do it across hundreds of stocks?

Then if we had a research on hundreds of stocks in our SQL database could we query the investment cases and narratives to find the best ideas, or ‘see’ the market?

That’s what we created with YDX and it’s how we came up with the LNG Terminals idea in Qatar changes Everything.

The 5 step process

But first a learning from Stevie. Stevie has API access to Google Sheets, so YWR readers would often ask Stevie to build an earnings model on a stock idea. And what I would notice is these financial models were not very good. Stevie was trying to do too much at once. Stevie was trying to research the company, build a model and come up with forecasts all at once in under 10 minutes.

For YDX I decided to break the research up into automated steps.

Step 1: Pull 5 years of historical financial data from the SEC (for US stocks). Just pull the data. Nothing else.

Step 2. Build a historical financial model (balance sheet, income statement and cash flows) based on a template I gave it. Make sure numbers like Net Income and Total Assets tie to the underlying line items. Just arrange the data into a financial model.

Step 3. Write a research report on the company. Analyse the products and services. Listen to the conference call. Just write a report. Be objective. Don't tell me whether you like the company or not, or whether I should buy it. Save it in the SQL database.

Step 4. Using the financial history (Step 2) and the research report (Step 3), build 3 years of financial forecasts. Save it in the SQL database.

Step 5. Using your financial forecasts (Step 4) and research report (Step 3) write an investment case. Save it in the SQL database.

Work through this 5 step pipeline at night.

The result is a research pipeline running at night as Stevie chews through 584 stocks. Currently, 302 have completed Stage 5 (the investment case). We have also already analysed 91 US energy stocks as part of building out our US energy view. So Stevie has analysed and modelled 393 stocks so far.

The output from each step is stored in our cloud based SQL database (I use Neon).

YDX Website

Finally, I built the YDX website to access Stevie’s research. You can read Stevie’s latest reports in Views or search by stock.

It started as a theoretical exercise, but I have to say it has turned out really well. I enjoy Stevie’s write ups. They’re pretty good.

Remember this is an AI research experiment. Not financial advice!!

You’ll see I built the YDX architecture around the possibility to have multiple authors. Because why not create different AI’s with different personas to analyse stocks? Same as the sell side. Stevie’s reports are all quite optimistic. Maybe we need to create a cynic and have it write 584 reports too. So we have the Bull and the Bear on every stock.

Caution: I know you know this, but treat Stevie and his write ups as overly bullish analysis which is slightly prone to exaggeration to get your attention. Use the mosaic theory. Trust but verify and look at both sides. And they are not financial advice!

Narrative Visualisation

At the end of the day we are investing in stories.

And if we have a database of stories we can start to do some interesting AI analysis at the index or sector level. For example, we can vectorise these investment cases then run a Principal Component Analysis (PCA) to display them visually.

I built an example of this for Stevie’s research on the US Energy Sector with each stock hot linked to the write up.

The PCA visually shows where the investment cases are ‘basically the same thing’. On the chart below you see Chevron and Exxon are very close to each other, which makes sense. Then towards the top you see a cluster of triangles which are all refinery stocks. It tells you this cluster has a differentiated investment case from everything else, which is true. Another example, is that the star shaped cluster (oilfield services) is closer in similarity to Chevron than it is to Valero. Which makes sense.

YWR Energy Universe Investment Case Visualisation.

Potential Next Steps?

Should we schedule Stevie to update these investment cases every 3 months, or after the next earnings report? Stevie would be like a sell side analysis updating his view after the results.

Should we create a Google Sheet to compare where Stevie’s EPS estimates differ from consensus?

Should we use this process to build bottom up views on indices and ETF’s?

The importance of being curious.

I wanted to share one final thought from working with AI on the US energy stocks for Qatar changes Everything, because it’s almost a contradiction.

I built the whole energy universe coverage with AI, but to some extent it didn’t help much. At the end I had 91 bull cases. They all seemed potentially interesting. Where do you start? Oil versus gas? Producers versus refiners? Water extraction? Pipelines?

In the end I use AI to explore a key seed idea I had.

The seed idea, which I knew was differentiated, which led me to focus on the LNG terminals as the key investment idea, originated from watching the Total Investor Day back in September 2025 and remembering the importance Patrick Pouyanne put on the new supply coming from the Qatar LNG projects. I remember how he said it and that he mentioned it again later in the Q&A. It stuck in my head. It was key to Patrick’s whole view on global LNG. Then when I saw the damage to the Qatar LNG from Iran, the idea clicked… this was going to change everything about global LNG.

Then I used the AI research universe to search for the best way to play it. The process worked well, because Stevie had already done background research on the entire sector.

So we can have all this AI analysis and coverage, but it helps to have a creative, original idea. The human still needs to connect the dots.

And the contradiction I have is that the connection came from sitting for an hour, potentially ‘wasting my time’ listening to Patrick Pouyanne. I had no idea if he would say anything interesting, but I take time to listen to Patrick because I like how he talks, and I alway often get good insights from him. Jensen Huang, Elon Musk, Jeffrey Sprecher, Gary Nagle, Andrea Orcel are all the same. It reminds me of that advice from when I was at my liberal arts college where we were encouraged to pick courses based on the teacher not the topic.

What I’m trying to say is I think we still need to go down rabbit holes, and ‘waste time’. We need to be curious. In fact that’s our key job. And we need to look out for great teachers and tune into that. Because that’s where the gold is.

AI can help with exploring rabbit holes and maybe make the process quicker, but you still have to take time to read things and make those intuitive, creative connections which AI doesn’t because our brains work differently (Primal Intelligence)

But then when we do have an insight AI can help us run with it quicker.

And if you are building out AI research for your firm and want to chat about it contact me at erik@ywr.world.

Have a good rest of the week!

Sorry, again for the 393 stock ideas.

Erik