Can transformers learn to read markets?

Project K.E.M. is an open experiment in multi-modal machine learning: market data, fundamentals, news, and inter-company relationships fused into one embedding per stock, with every model output tracked forward and published — including the results that don't flatter the model.

How the system works Explore the embedding space

Daily Model Output

Today's Signals

What the model flagged in its latest run — published as research output, not as recommendations. Quiet days are shown as quiet days.

Signal detail & methodology

Live, Realized Results

Measured Performance

Computed from realized close-to-close outcomes — no marketing figures, no simulated numbers.

Loading live performance…

View Detailed Performance

The Architecture

Four modalities, one embedding

Each stock is represented by fusing four separately-encoded views of it. The full two-arm pipeline — novelty detection and probabilistic forecasting — is diagrammed on the Technology page.

Market Data

Processes price patterns, volume, and technical indicators to identify significant market trends.

Fundamental Data

Analyzes financial statements, earnings reports, and valuation metrics to assess company health.

News & Social

Processes news articles, social media, and analyst reports to capture market sentiment.

Relational Data

Maps connections between companies, supply chains, and industry relationships.

See the full architecture

Questions

Frequently Asked Questions

Get answers to common questions about Project K.E.M.

Frequently Asked Questions

: Project K.E.M. (Knowledge Embedding Modalities) is an open academic research dashboard. It publishes the live, forward-tracked predictions of a multi-modal machine-learning system that fuses textual, technical, fundamental, and graph representations of a stock into a single embedding. It is a research artifact, not a trading advisory or brokerage product.
: The platform uses a two-arm architecture. Arm 1 (Novelty Detection) measures how far a stock's current fused embedding deviates from its own historical baseline. Arm 2 (Probabilistic Forecasting) is a supervised transformer that estimates the probability of a positive or negative price movement across three timescales. A concordance step reports where the two arms agree. See the Guide and Technology pages for details.
: Every metric on this site is computed from realized, close-to-close market outcomes after a model's release date — never backfilled or hand-picked to look good. Predictions are timestamped and tracked forward from release to avoid look-ahead bias. The models are experimental and unvalidated; past measured behavior is not evidence of future results and nothing here is a performance guarantee.
: Three: next_day (1 trading day), short_term (5 trading days), and long_term (20 trading days). Each is evaluated independently, so a model can be informative on one horizon and uninformative on another.
: No. Project K.E.M. is not a registered investment adviser and does not provide personalized investment advice. Its outputs are the result of a mathematical model shared for research and educational transparency, and should never be the sole basis for any investment decision. Always do your own research and consult a qualified financial professional.
: The current research universe is a fixed set of U.S. equities listed on the NYSE and NASDAQ. Coverage is intentionally limited so that model behavior can be studied and reproduced rather than marketed as a broad product.

Browse the research universe

62 U.S. large-cap stocks, each with its own model-output history — novelty, forecasts, and concordance per trading day. Deliberately small, so behavior can be studied rather than skimmed.

Go to Stock Directory

IMPORTANT DISCLAIMER:

The information provided by Project K.E.M. is for informational purposes only and does not constitute financial advice, investment recommendations, or an offer to buy or sell securities. The signals generated are the result of a mathematical model and should not be the sole basis for any investment decision.

Past performance is not indicative of future results. All investments involve risk and the potential for loss. Users should conduct their own research and consult with qualified financial professionals before making investment decisions.

Project K.E.M. is not a registered investment advisor and does not provide personalized financial advice.