I turned my own investment research methodology into an AI research engine.

You input a piece of code, and it runs for 23 minutes, calling 9 data sources and using 25 tools. But the point isn’t speed.
The point is—once it finishes these four things, it only then dares to speak:
1｜First assess whether it lives or dies, not price first.
Is this crisis temporary or permanent? Is it an emotional mispricing, or a structural decline? If it can’t clear this step, the research ends on the spot.
2｜Valuation is given only as a range, never a target price.
More than a dozen methods cross-check: DCF, replacement cost, historical percentiles, reverse DCF. A target price precise to the decimal place is false precision—we only say how “cheap” it gets, and what zone it reaches.
3｜Before reaching a conclusion, attack your own thesis first.
Adversarial red-teaming: go line by line and challenge your own numbers. Any data that could flip the conclusion must be traced back to the original disclosure (SEC / HKEX); secondary retellings aren’t accepted.
4｜Every number must be traceable back.
An automated verification gate aligns every number in the report to the original evidence. Every number you see has a source attached to it. If the data isn’t enough, it says plainly: “the data is thin and unreliable”—instead of inventing a nice-sounding figure.
In the end, it gives only two conclusions: buy, or veto. No “wait and see.”
And it must also spell out a falsification line: what signal appearing would overturn that judgment. It doesn’t generate buy/sell instructions.
It judges what counts as “cheap” and what counts as “catalysts”—the decision is yours. Feel free to criticize!