Loading
Loading
StockBrain builds living AI research models that learn from filings, earnings, metrics, and market evidence over time.
Not a stock picker, not a prediction engine, not a trading system. A research memory that reads primary documents, holds a structured view, and shows its work.
The thesis rests on data-centre compute demand outrunning supply for longer than the market is pricing, with CUDA as the switching cost that keeps share.
The thesis rests on data-centre compute demand outrunning supply for longer than the market is pricing, with CUDA as the switching cost that keeps share.
Illustrative thesis graph. A live brain draws this from its own evidence, and every node links to the filings behind it.
One loop, run every time a new document arrives. Nothing enters a thesis without a document behind it, and every step leaves a record.
A filing, earnings release, or dataset is retrieved whole, hashed, and stored with the URL it came from.
Verbatim spans become structured facts — metric, value, period, quote. Nothing is paraphrased.
Each fact is read against the thesis it already holds and classified: supporting, contradicting, or genuinely new.
The evidence lands on a concept in the graph, moving that node’s status and the weight of the links around it.
After a material event the brain audits its prior view: what held, what failed, what it never considered.
The revised view is sealed as an immutable version, hash-chained to the one before it.
Five systems, each doing one job. Together they are the difference between a model that answers a question and a model that keeps a position.
Each brain is a persistent model of one company. Open one to read its current thesis, the evidence on both sides of it, and every version it has held.
No evidence classified yet.
No evidence classified yet.
No evidence classified yet.
This is what the product actually does with a filing. A sentence becomes a structured fact, the fact becomes evidence with a stated reason, and the evidence lands on a named concept in the graph — with the document still attached at every step.
“Data Center revenue was $89.0 billion, up 117% from a year ago and up 18% sequentially, driven by the ramp of our Blackwell Ultra infrastructure.”
Part I › Item 2. Management’s Discussion and Analysis › Second Quarter of Fiscal Year 2027 Summary
data_center_revenue$89.0BQ2 FY2027 · +117% y/yExtracted with the sentence above stored verbatim alongside it. A fact never exists without the span it came from.
Read against the thesis already on file: acceleration on a base this large is consistent with demand outrunning supply, which is the claim the thesis rests on. The reasoning is stored, not just the label.
Attached to the concept it bears on — the same node shown in the graph above. Its status holds, and the link from AI Infrastructure Demand carries more weight than it did last quarter.
The revised graph is sealed and hash-chained to its predecessor. Three months from now you can open this version, read what the brain believed, and see the single piece of evidence that moved it.
Quote, filing metadata and link are verbatim from the document above. The classification shown is an illustration of the pipeline’s output format — on a live brain it is produced by the evidence engine and carries its own timestamp.
StockBrain does not pick stocks, forecast prices, or place trades. It reads primary documents, holds a structured view of each company, and shows you exactly what changed and why — with the source still attached.
StockBrain publishes model-generated research interpretations, not investment advice, recommendations, or forecasts. Evidence is linked to primary sources so every claim can be checked.