The Dreame Universe—Investment Due Diligence Written by AI

How do you get AI to turn out the kind of company due diligence only a veteran investor could plausibly produce? That’s what this episode is about.

Research with AI is already easy

I’ve mentioned this in earlier videos: doing research with Claude Code is already dead simple—it can fire off a web search on its own, rank sources by how trustworthy each one is, organize everything logically, and finally present the report as a web page.

But there’s one gap in that process: a lot of information simply isn’t public—Claude Code genuinely can’t find it online.

So what do you do?

That’s when you need a professional data provider behind you.

Which platforms in China do corporate information well and are also agent-friendly? I asked Claude Code, and it pointed me to Qichacha’s agent platform.

The workflow: identity verification → MCP config → Claude Code

Open the Qichacha agent platform and start with identity verification—the interface is currently only open to companies. But if you have a company, it’s easy: about half an hour and you’re through their company verification, with MCP access in hand.

All you have to do is hand that config file to Claude Code, and you can go research companies as recklessly as you like, on the free credits they give you.

Once Claude Code has the config, you tell it:

• what purpose I’m doing this for
• which company I want researched
• which angles to research
• what else I need Claude Code to fill in using its own search
• and what kind of report I want at the end

Once you’ve told Claude Code all of that, roughly fifteen to twenty minutes later you get back a very detailed piece of due diligence.

Demo: the Dreame Universe

About the Dreame Universe report I made—Boss Yu may not be posting a hundred updates a day online anymore, but I’m sticking to the data here, with none of my own opinions in it.

This piece of due diligence is entirely AI-made. I didn’t change a thing; it came back in one pass. Here’s the level of detail you get:

Overview

It opens by telling us how Dreame grew, in eight years, into an ecosystem of nearly a thousand companies, leveraging 67 industrial funds and tens of billions of yuan in capital. Then how much of that comes from government LPs, and how much of the revenue comes from overseas—an overview first.

The controlling shareholder + ownership structure

Next comes the controlling shareholder’s own trajectory, and the mechanisms he uses to hold his stakes, laid out plainly.

Further down, which companies hold the controlling stakes in the Dreame entity and what the percentages are—built out of public, credible reporting plus Qichacha’s data into a clean pie chart.

The five strategic pillars

Further down still, the five strategies Dreame itself talks about—how it does appliances, how it does cars, how it does humanoid robots and embodied intelligence, plus satellites, chips, and the overseas entities. Which companies are involved in the capital maneuvering, and through what holding structures—all of it laid out clearly.

The relationship network map

It puts Yu Hao himself at the center node of the network and draws a relationship graph. We can zoom out and see every related company, including the earliest co-founders and the people involved—the whole chart is there. Holding relationships, drilled down through the layers.

Capital flows

Further down is the capital injection. There’s the Jiamei Packaging situation—again, no opinions from me—and the whole capital-flow diagram is crystal clear. You can also see how the money subscribed by local governments flows back into Dreame’s ecosystem—extremely clear. Remarkable.

Overseas + timeline

Then which overseas entities exist, which key events they’re tangled up in, market share by region, and the patent footprint.

Then the company’s development timeline—how eight years unfolded, milestone by milestone, which companies completed investments in it, what the co-founders are working on now and how they relate to the parent company—forming an ecosystem jigsaw.

Risk signals + future trends

The risk signals that have already materialized—governance-structure problems, valuation problems, the backdoor-listing problem, plus the recent public sentiment issue around Yu Hao running his own creator channels.

The current strategy—inferred from public statements and PR copy, and reverse-engineered from the moves he has actually made, including capital leverage and the pattern of executives leaving to form the ecosystem.

It also projects the key events that could happen in the next one to three years—sorted by time window and probability. Filing with HKEX, expanding the government-fund program, the automotive business advancing further, and various subsidiary movements.

Risks that haven’t happened but could, sorted from high to low. I won’t go through them all here—download the page from the blog and read it yourself.

Limitations—it knows what it doesn’t know

The report’s limitations: which pieces of information it couldn’t definitively verify—it flags every one of them. Then where each piece of information came from and how credible that source is—listed-company filings, official overseas data, financial media, industry research, every source marked.

It even goes as far as attaching a small superscript marker to every claim above, saying where it came from, and you can click through to the original.

The room this opens up

The first time I saw this, I was completely floored.

And I’m also grateful I was born a bit earlier—I’m just an old-timer, born in ‘85. If I’d been born in the 2000s, I might not even get the chance to do the grunt work, the information-gathering, for the big guys—that chance would already be gone. It’s gotten that good.

And this is only a demo. If you can understand a company end to end like this, plus real-time news search, plus Claude Code—then for investing, the room this opens up is enormous.

Source: EP0023_audio.mp3 · ASR model gemini-2.5-pro (chunked parallel) · full text of the 6:51 original recording

[00:00] How do you use AI to produce the kind of company due diligence only a veteran investor could pull off? That’s what we’re talking about today. I’ve mentioned in earlier videos that doing research with Claude Code is already dead simple. It can fire up web search on its own, sort sources by how much you can trust them, and lay it all out in a way that actually holds together. And at the end it’ll hand you the report as a web page. But the thing missing from that whole process—

[00:27] the gap is this: a lot of information simply isn’t public. Which means our Claude Code can’t actually find it online. So what do you do? In that situation you need a professional data provider backing you up. In China, right now—I actually asked Claude Code about this. I said, which platforms in China are good at company information and friendly to agents? And in the end Claude Code recommended Qichacha’s agent platform.

[00:53] So I opened up the Qichacha agent platform, and the first thing I did was real-name verification, because their API is only open to companies right now. But as long as you have a company it’s easy—about half an hour and you can get through their company real-name verification and get MCP access. Then you just copy that config file over to Claude Code, and we can pretty much go nuts burning through the free credits they give you doing company—

[01:19] research. So once you’ve handed that to Claude Code, you just tell it: here’s what I’m trying to do, here’s the company I want to look into, here are the angles I want it looked at from, and here’s the extra stuff I want Claude Code to fill in with its own search. And here’s the kind of report I want at the end. Once you’ve told Claude Code all of that, in maybe fifteen to twenty minutes you get back an extremely detailed due diligence report. So you can see the one I did on the Dreame universe.

[01:44] Now, Mr. Yu may not be posting 100 times a day online anymore, but in this episode we’re only going through the hard data, no personal opinions. So looking at this due diligence that AI produced entirely on its own—I didn’t change a thing, one pass, didn’t touch it. How detailed does it get? First it tells us that in eight years it grew into an ecosystem network of nearly a thousand companies, right? Then how many funds it levered up—67

[02:10] industrial funds, with tens of billions of yuan in capital. Then what share of the LPs are government, including what share of revenue is overseas. It leads with an overview, right? Then, right after, the story of how the actual controlling person came up, and how he holds his stakes—laid out crystal clear. Then further down, which companies are the controlling persons of the Dreame entity, and what each one’s stake is. Through public, credible reporting,

[02:35] plus Qichacha’s data, it can also produce a nice pie chart like this. Then further down, the five big strategies Dreame officially talks about—how they do appliances, how they do cars, how they do human… humanoid robots, embodied intelligence, and then satellites and chips, plus their overseas entities, and which companies are wrapped up in their capital moves. And then how, through what—

[03:00] what kind of setup they use to take controlling stakes—it’s all laid out super clearly in here. Then it also puts Yu Hao himself at the center of the network and starts drawing this relationship graph. We can zoom out—every related company, including his earliest co-founders, the people involved. The charts for all of these companies are in here, you can see all of it: who controls who, the ownership drill-down. Then further down, the capital injections.

[03:26] Like I mentioned earlier, the Jiamei Packaging thing—again, we’re leaving out opinions. We can see an extremely clear diagram of the whole capital flow. And you can also see how the money subscribed by local governments flows back into the Dreame ecosystem. Absolutely crystal clear. And then which overseas entities it has, and which key events those touch.

[03:52] Its market share by state and city, plus its patent footprint. Then the company’s development timeline—how it grew over eight years, milestone by milestone, which companies invested in it, including what businesses its co-founders are running now, and their relationship with the parent company—forming an ecosystem jigsaw, a jigsaw puzzle.

[04:18] Then the risk signals that have already surfaced, right—there’s the govern… structural governance problem, then there’s the valuation problem, then there’s the backdoor-listing problem, then there’s the internal—including the recent thing with Yu Hao running his own creator account and the public-sentiment problem around that. And then the current growth strategy—it pieces that together from his own public remarks,

[04:43] plus a few PR pieces, and it works backwards from the moves he’s actually made, to get at his core strategies—including the capital leverage, including the way executives leave to build out the ecosystem, that whole pattern, right. And then it also forecasts the key events that could happen in the next one to three years, ranked by time window and probability. For example, it might file with the Hong Kong—

[05:08] Stock Exchange, and it might scale up the government funds, and the car business, the car business might get further off the ground, including some of the moves by its subsidiaries. Risks that haven’t happened yet but could, ranked high to low. Possible risks—I won’t go on about that part, you can go download the web page from the blog and read it yourself.

[05:34] Including some of the things the founder has said publicly—all in here. And then the limitations of this report, right—which parts of it, which pieces of information it couldn’t nail down for sure. It flags all of that for you. And then which sources it used and how credible each one is—for instance, listed-company filings, then overseas official data, then financial media, plus industry research.

[06:00] Every single source gets marked in here. And it even goes as far as putting a little superscript on every piece of information up there telling you where it came from, and you click it and you see the original text. So at this point—honestly, the first time I saw this I was completely blown away. And I’m also glad I was born a bit earlier. I’m a post-‘85 guy, born in ‘85, an old-timer. If I were born after 2000, I might not even get the chance to be the gofer for—

[06:25] some big shot, doing that kind of information-gathering grunt work. It’s already this good. And this is just a demo. If we have this capability—this ability to fully understand a company, plus real-time news search—then for investing, if you know how to use Claude Code and you add something like this on top, there’s a ton of room to run with it.