Approach
Six asset classes, three tiers, one risk frame.
AYVID runs a multi-asset strategy on a three-tier architecture. The system was designed under the constraint that any subsystem can fail and the rest must degrade gracefully — no single agent, model, or data feed gets to take the firm down.
Track A · Index futures & options
Direction-of-shock trades on global indices.
We trade futures and short-dated options across roughly twenty index venues — S&P 500, Nasdaq 100, FTSE 100, DAX, Nikkei 225, Hang Seng, KOSPI 200, Nifty 50, Sensex, ASX 200, and others — depending on which markets are open when a shock prints.
Sizing is conviction-weighted with hard −1R stops and per-index cooldowns. We never compound losses inside the same shock window.
Track B · Equities
Causally-implied beneficiaries and victims.
When a shock breaks, the graph proposes the names that should move — direct exposure, secondary supplier dependencies, geographic chokepoints. We screen these against liquidity and cost-to-trade in real time, and only the survivors enter the equity book.
The universe is ~50+ tickers, dynamically rebuilt every cycle. There is no fixed watchlist; the engine selects what to watch from the graph itself.
Multi-asset expression
One shock, six asset classes.
A single geopolitical event simultaneously routes to equities, index F&O, FX, commodities, crypto, ETFs, and bonds. The graph decides which asset classes are structurally exposed — and the execution engine sizes each expression independently with per-asset-class cost models, slippage assumptions, and exposure caps.
FX macro trends, commodity supply shocks, crypto momentum, ETF sector rotation, and rate-curve trades all fire from the same causal thesis. Multi-asset expression is how we capture the full consequence surface of every shock the graph detects.
Risk frame
The kill switch is independent.
An out-of-process C++ watchdog monitors equity, drawdown, and order flow every tick. If equity breaches the daily loss budget, or if order flow looks anomalous (rate, sign, or imbalance), the watchdog flattens the book without consulting Python — sub-millisecond.
This is the single most important architectural choice we made. It says explicitly that we don't trust our own LLMs to behave under stress.
Want to dig deeper?