01 · Non-stationarity
The core problem.
Statistical properties of markets shift in minutes. Models trained on yesterday's regime fail today's. We treat regime adaptation as a first-class engineering problem.
A research project by Blocksoft GmbH
A Foundation Model for financial intelligence.
We train AI systems that learn market structure — across asset classes, across decades — and deploy them as institutional infrastructure.
Multi-modal
Numerical · Textual · Graph-structured
Multi-decade
22 years · survivorship-bias-free
Multi-market
Equities · FI · FX · Commodities · Digital
Research validated in live production
Currently deployed
First deployment surface · non-custodial yield engine
Live-data validation
365-day backtest · 2.70 Sharpe · 3.01% Max DD · 19.81% APY · 83.2% positive days
Crash-tested in production
Live engine continued +2.19% APY through BTC −52%
A research project by Blocksoft GmbH
The problem we study
§ 01
Financial markets are non-stationary. Current models aren't.
Most quantitative models assume the future resembles the past in structured ways. Markets disagree. Regimes shift. Correlations break. The gap between research and production is where institutional capital gets hurt.
02 · Calibration
Matters more than accuracy.
An uncalibrated prediction is dangerous regardless of accuracy. Institutional risk needs honest confidence intervals — and a system that reports when it does not know.
03 · Validation
Must be honest.
Walk-forward expanding-window cross-validation across decades — including 2008, 2020, 2022, 2026 — is the only honest way to measure financial-model performance.
Our research
§ 02
Trained on the market. Validated against decades. Deployed under your policy.
A Foundation Model for financial intelligence — multi-modal architecture, trained self-supervised on multi-decade survivorship-bias-free data across asset classes. Calibrated forecasts. Regime-aware. Auditable.
01
Trained on the full market.
22 years of survivorship-bias-free data — equities, fixed income, FX, commodities, digital assets. Self-supervised pretraining + walk-forward fine-tuning. Validated through every regime since 2007.
02
Calibrated uncertainty.
Every output ships with a confidence interval. The model reports when it does not know. A non-optional system property, not a research curiosity.
03
Online adaptation, under policy.
Adapts to regime shifts in real time within institutional policy guardrails. Every adaptation logged, auditable, reversible.
Compute & training
Pretraining, walk-forward fine-tuning, and ablation studies require ~5,000–10,000 GPU/TPU-hours over 12 months. Compute partnerships and credit programs welcome.
Request research brief (NDA) →
Methodology under NDA for qualified investors, institutional partners, research collaborators.
Live in production
Beta · live
§ 03
The yield engine is live. The data is real.
Our first deployment surface — a non-custodial yield-allocation engine running in beta at app.yielz.ai. Performance is verifiable.
What this is, what it isn't.
The numbers below describe our currently deployed yield engine — the first product surface of our research. It is not the Foundation Model we are training (that work is described in § 02 Our Research). These metrics validate that the research-led approach already produces a working institutional product.
19.81%
APY
365-day backtest · Dec '24–Dec '25
2.70
Sharpe
institutional-grade
3.01%
Max DD
across backtest
83.2%
Positive days
stability signal
+2.19%
Feb '26 · live
BTC −52% crash event
Under the hood
Three engines turn model output into execution.
Opportunity Radar
Continuous discovery
Surfaces 15,000+ yield opportunities across 30+ networks — on-chain today, money markets and structured credit on roadmap. Policy-filtered.
Risk X-Ray
Multi-factor scoring
Per-opportunity scoring on counterparty, contract risk, oracle dependence, liquidity depth, and hundreds more. Explainable reasons. No black boxes.
Compliance-Aware Router
Non-custodial execution
Non-custodial allocation under policy. Automated rebalancing. Dual-trigger emergency exits. Multi-sig / MPC / custodian compatible.
Beta · qualified participants · capped exposures · on-chain venues only. Cross-market expansion (money markets, structured credit, tokenized assets) on the institutional roadmap. Past performance does not guarantee future results.
Security & compliance
§ 04
Designed by cybersecurity experts. Operated to institutional standards.
01
Verified & monitored
Independent contract and venue audits. Continuous monitoring detects exploits, depegs, and venue anomalies. Dual-trigger automated evacuation — TVL-shock + stablecoin-depeg detection.
02
Non-custodial
You hold the keys. Withdraw instantly. No lock-ups, no gates, no counterparty risk on yielz.
03
Compliance framework
SOC 2 in progress. ISO 27001 / 23894 / 42001 aligned. GDPR / CCPA ready. MiCA-aligned workflows. EU AI Act framework experience.
Who we serve
§ 05
Built for institutional capital — wherever it is.
01 · Priority audience
Asset Managers, Hedge Funds & Wealth Platforms
Need
Clients want cross-market yield products. In-house quant infra takes years.
Fix
White-label our deployment engine. Embed model-driven allocation under your brand.
Win
New revenue · client retention · differentiation.
02
Corporate Treasuries & Family Offices
Need
€100M+ cash earning 2% while inflation runs 3–4%. Teams can't safely cover dozens of surfaces.
Fix
AI-driven yield optimization with full compliance, audit trails, emergency exits.
Win
€5–10M+ per €100M deployed · zero loss of custody.
03 · Current beta deployment
Crypto-Native Treasuries
Need
Stablecoin treasuries, market-maker reserves, protocol treasuries sitting idle.
Fix
Same model, same discipline, currently optimized for on-chain surfaces.
Win
Institutional discipline for crypto balance sheets.
Questions we get
§ 09
Common questions.
The Foundation Model is built for the full investable universe — equities, fixed income, FX, commodities, and digital assets. Multi-modal architecture across numerical series, filings, news flow, and graph-structured data. Trained on ~22 years of survivorship-bias-free data across all asset classes.
No. The yield engine at app.yielz.ai is the first deployment surface of the Foundation Model. Blockchain yield was the fastest market to validate the research approach in production. Cross-market expansion to traditional money markets, structured credit, tokenized treasuries, FX overlays, and equity allocation is on the institutional roadmap.
We are not an aggregator — we don't run pools. We are not a hedge fund — we don't take discretionary positions. We operate proprietary AI systems that score opportunities, anticipate regime shifts, and adapt in real time within your policy. Detailed methodology is available under NDA to qualified institutional partners.
No. You hold the keys. We never touch funds. Non-custodial architecture is foundational — and is preserved across all asset classes on the roadmap.
Work with us
§ 06
Deploy the engine on your treasury.
Institutional onboarding from €100K. Full compliance, audit trail, 24-hour timeline.
partnership inquiries: [email protected]