HELIOS
Live on dYdX v4 · Hyperliquid

Owen Hobbs

Trading Systems · AI Architecture · Scalable Design

Ex-Head of Trading & Product Platforms · CoinAlpha

I built a three-layer AI agent platform that manages live algorithmic trading — from autonomous research through deployment to 24/7 monitoring.

Return

24-mo backtest

Sharpe

Risk-adjusted

Max Drawdown

Risk control

Win Rate

Consistency

Results

ProScore2 · Mar 2024 – Mar 2026 · IS/OOS validated

Return

Win Rate

of trades profitable

Sharpe Ratio

risk-adjusted return

Max Drawdown

worst peak-to-trough drop

Loading...
BTC-USD 15m | 24-month backtest | IS/OOS validatedFull dashboard

What I Built

Autonomous Research

Agents test trading hypotheses through iterative experimentation — baseline measurement, parameter optimization, walk-forward validation. Budget-constrained with quality gates on every action.

IS/OOS validated · Monte Carlo · budget-constrained

Multi-Exchange Trading

Live on dYdX v4 and Hyperliquid with automated risk controls, circuit breakers, and a webhook pipeline feeding real trades to this dashboard.

dYdX v4 · Hyperliquid · automated risk controls

Deterministic-First Monitoring

Three-layer system: free deterministic checks do the heavy lifting, LLM analysts handle anomalies on-demand, rule-based router orchestrates. 24/7 autonomous with crash recovery.

93% cost reduction · 24/7 autonomous · crash recovery

The Pivot

the architecture decision

Built 6 Claude-powered agents for 24/7 monitoring. Then realized: 95%+ of monitoring cycles find nothing wrong. Paying an LLM to confirm "everything is fine" is wasteful. Rebuilt as a three-layer architecture — deterministic checks first (free, instant), LLM analysts only when anomalies require reasoning. Same coverage, 93% lower cost.

What It Does Today

The research agent takes a trading hypothesis and autonomously tests it — from baseline measurement through parameter optimization to IS/OOS walk-forward validation.

research-agent
0/14
$ helios research --hypothesis "mean-reversion with funding rate filter"
[PLAN] Evaluating hypothesis: mean-reversion with synthetic funding rate on BTC-USD
BACKTESTEstablishing baseline performance on default parameters
→ 100 trades | Sharpe 1.28 | Win rate 41.0% | Max DD -26.4%
OPTIMIZERunning parameter sweep across funding rate thresholds
→ 162 combinations | 5.6% pass rate | Best Sharpe 1.78
VALIDATEWalk-forward IS/OOS validation on optimal configuration
→ 54 trades | Sharpe 1.78 | Win rate 51.9% | Max DD -16.0%
CONCLUDEHypothesis confirmed — funding rate filter improves risk-adjusted returns
✓ VERDICT: Deploy to live monitoring (Phase B)
Budget: 7 rounds | 8 LLM calls | 5 optimization sweeps