The AI’s purpose is to determine the best possible strategy(s) a vault should be in based on market conditions and a vault’s risk profile. The best possible strategy(s) is determined by selecting the best predicted strategy returns which are given by the best prediction models, which in turn are generated by training prediction algorithms.
Prediction algorithm training has to be done on a vault-basis, that is to say every vault has its own training workflow, providing tailored prediction models based on the Vault’s desired configuration, which includes:
- AI Engine Fitness Function. (“Maximize returns”, “Minimize Drawdown”, etc …)
- Investment universe. (Pools and Tokens)
- Risk Profile
Training of these algorithms is a continuously updating process that generates and updates models as the market environment changes.
The AI Engine is implemented in a custom developed python library Opty.py which will be open-sourced and shared with the community. Key features include:
- Data Ingestion Pipeline