Train Agents
That Trade
A metaclass-driven, registry-first RL framework. Hash-locked RLExperimentSpec, RLRuntime, and Iceberg trajectory store for deterministic, replayable training and evaluation.
RLRuntime
Hash-locked RLExperimentSpec ensures reproducibility. Every experiment is versioned in PostgreSQL. Trajectories are stored in Iceberg for forensic replay.
Pre-Built RL Agents
Choose from 12+ industry-standard RL architectures including EIIE, DeepTrader, Investor-Imitator, ETEO, and OPD. Optimized for low-latency financial environments.
Market Dynamics Modeling
Slice-and-merge regime labeling. Train agents that are aware of market conditions and can adapt their strategy to changing regimes.
PRUDEX Evaluation
17 independent measures across profitability, risk, diversity, execution, and explainability. Moving beyond Sharpe ratio for professional trading.
Weight-Centric Pipeline
FinRL-X inspired 4-stage pipeline: Feature selection -> Alpha generation -> Target weights -> Risk overlay. Robust portfolio-level decision making.
The PRUDEX Evaluation Framework
Moving beyond Sharpe ratio. Our Phase 9 PRUDEX-Compass framework provides 17 independent measures and 5 advanced visualizations to truly understand agent behavior before deploying capital.
Iceberg Trajectory Store
Every step, observation, and reward is persisted to an Iceberg warehouse for forensic analysis and replay.
Weight-Centric Pipeline
FinRL-X inspired pipeline (f_S → f_A → f_T → f_R) for robust portfolio-level decision making.
Pre-built Agents
- EIIE (Ensemble of Identical Independent Experts)
- DeepTrader (Asset scoring + risk control)
- Investor Imitator (Inverse RL)
- DeepScalper (HFT execution)
- PPO In-house (Optimized for low-latency)
- FinAgent (LLM-hybrid adapter)