SageResearchAI

Sage Research builds vision transformer models for financial markets, uncovering hidden structure beyond traditional indicators. Institutional inquiries: DM.

[SageResearchAI @SageResearch_AI]

https://t.co/o02oFzWCxQ

[SageResearchAI @SageResearch_AI]

Our work. We extract latent structure from trading models -- yours or ours -- for institutional clients.

See more alpha...

→ https://t.co/F7Cu31p8Xt

[AlphaVision.market @AlphaVision_Mkt]

We train AI vision models to predict the stock market.

🧵Video of 3D structure + thread ↓ ↓ ↓

[World of Science @Science_TechTV]

Two brain cells are sensing each other. And trying to connect:

[SageResearchAI @SageResearch_AI]

Seven Years of Mag7...
According to our AI Vision Model

This is what 7 years of Mag7 looks like to our AI vision model.

Every dot is one of our model's weekly predictions on the Mag7 stocks (AAPL, AMZN, GOOG, META, MSFT, NVDA, TSLA).

2,578 dots in total. Color = year. Dark purple is 2019. Yellow is 2026.

Our model has never been told what year it is.

It only sees heat maps of market data -- prices, interest rates, volatility, sentiment, etc.

But the years group themselves anyway.

The model can tell 2019 markets apart from 2026 markets just by looking at the data.

That means it has learned to recognize different market "eras" -- like COVID, the zero-rates boom, and the rate-hike cycle -- without anyone teaching it.

And in 2026 it's seeing a "regime shift".

[SageResearchAI @SageResearch_AI]

7/7

6. Year

The headline. A clean purple-to-yellow gradient. The model independently rediscovered that 2018–2020 and 2024–2026 are different market regimes — without ever being told what year it was.

[SageResearchAI @SageResearch_AI]

What does a Vision Transformer trained on market data actually "think about"?

We took every Mag7 prediction AlphaVision has ever made, pulled the model's internal embeddings, and projected them to 3D — colored by 6 different concepts.

Thread below...
1/7 🧵

[SageResearchAI @SageResearch_AI]

6/7

5. Model confidence (P(UP))

Most predictions sit near the indecisive middle. The high-conviction band is small but visible — that's where the trade alerts come from.

[SageResearchAI @SageResearch_AI]

That's because markets behave more like weather systems than spreadsheets.

Sector rotations.
Volatility clusters.
Liquidity waves.
Panic cascades.

Our project is to turn market data into satellite imagery for AI.

So the model can see what traders miss.

[SageResearchAI @SageResearch_AI]

“This shape often appears before large asset moves.”

Traditional investing looks at indicators.

AI Vision looks at structure.

Because the most profitable signal may not be in any one number.

It may be in the shape they form together.

[SageResearchAI @SageResearch_AI]

Vision transformers see something different.

They read market data as image-like heatmaps.

Not:

“RSI = 62”

But:

“Across billions of time-and-variable relationships, which patterns appeared before major moves?”

[SageResearchAI @SageResearch_AI]

Today, we launch AlphaVision.

Sage Research built AlphaVision around a simple idea:

Markets contain structure.

And now, AI vision can help bring that structure into view.

After 18 months in stealth, research, iteration, and serious compute, our first self-directed product is now live.

https://t.co/ALxs9RLang

[SageResearchAI @SageResearch_AI]

Tomorrow, we launch AlphaVision.

After 18 months in stealth, research, iteration, and serious compute, Sage Research is stepping into public view with our first self-directed AI vision product.

We built AlphaVision around a simple idea:

Markets contain structure.

And now, AI can help bring that structure into view.

Launching tomorrow.

[SageResearchAI @SageResearch_AI]

The Hidden Structure of Winning Trades

Over the last 18 months, one of the more interesting things we discovered while training AI vision models on market data:

Winning trades often unfold in two phases.

The first is the setup.

This is the part of the pattern where the underlying market structure begins to form.

Not the move itself — the conditions behind it.

Then comes the trigger.

This is the point where that structure starts to resolve into action.

In other words, the model isn’t just learning to recognize the trade.

It’s learning to recognize both the environment that tends to produce winning (or losing) trades.

In short: AI vision models CAN uncover the hidden structure to winning trades.

Until now, most tools in finance focus on isolated signals.

But AI vision can learn the broader structure surrounding a trade — including how that structure develops over time.

[SageResearchAI @SageResearch_AI]

For the past 18 months, Sage Research has been building in stealth.

Today, we’re sharing a first look at who we are, why we built AlphaVision, and the idea behind it:

Markets contain structure that traders can often recognize more clearly through visual patterns than through traditional indicator-based tools.

Watch here: https://t.co/TLxytu8xh8

AlphaVision launches April 28.

[SageResearchAI @SageResearch_AI]

A new way to see market structure launches April 28.

For the past 18 months, Sage Research has worked in stealth to build a new tool for traders.

We built it through research, iteration, and relentless compute, and now we’re ready to step into public view.

Our thesis is simple:

Markets contain structure that traders can often recognize more clearly through visual patterns than through traditional indicator-based tools.

On April 28, we’ll introduce AlphaVision — our first retail AI vision product built from that work.