The problem Persona Finance AI (PFAI) solves

Limitations of existing AI trading systems

Rigid strategies, static models based on historical data are difficult to adapt to market mutations, and parameter adjustments lag behind market fluctuation cycles

Challenges faced by users

Difficulty in understanding strategies . need to understand underlying mechanisms such as AMM/Pooling, and parameter tuning relies on professional quantitative knowledge.
Unsustainable returns . risk of impermanent loss, and strategy homogeneity leading to declining returns.
Lack of personalization . general strategies cannot match individual risk preferences, and lack of user behavior feedback loop.
No growth feedback . unable to quantify strategy evolution trajectory, steep learning curve and no visualization tools.

Persona Finance AI (PFAI) is the first DeFi protocol that combines trading strategy personalization with on-chain behavior evolution. By creating an on-chain trading personality with self-evolution capabilities (using ERC-6551 NFT as a carrier), It is market makers, algo traders, liquidity miners, quant developers, bot traders, researchers, cross-chain staking, self-evolving AI Agent, etc...

PFAI solves three core problems:

1. Strategy rigidity - traditional AI trading strategies cannot adapt to market dynamics
2. User threshold - DeFi complex strategies are difficult for ordinary users to master
3. Value capture - high-quality strategies are difficult to share value

PFAI achieves breakthroughs through three innovative modules:

• Dynamic personality casting: user risk profile → initial personality NFT generation
• Behavioral evolution engine: on-chain transactions → real-time DNA evolution
• Social financial ecology: personality leasing + trading + influence tokens

Challenges we ran into

One of the biggest problems we encountered was the implementation of the evolutionary algorithm. After we used AI to generate basic strategies based on user preferences, we needed to evolve its strategy based on each subsequent transaction of the user. In fact, we have been constantly trying to determine a reasonable range based on what content to adjust, how large the adjustment range is, or other factors. In the end, we completed a relatively rough evolutionary algorithm, but time was really too short.
Another problem is that we tried to make a model based on Eliza OS. Due to insufficient research, the agent we made did not achieve the desired effect.
We will continue to complete this project.