Nilay Kumar - EtherFi Strategy Arena

Inspiration

I’ve always wanted to get into crypto, but I never really understood how staking and yield strategies actually worked behind the scenes. EtherFi’s approach to restaking caught my attention, and I wanted to build something that helps visualize how different strategies can produce different returns depending on risk, gas prices, and leverage. This project became my way to both learn and simulate those dynamics in a more intuitive way.

What it does

It's an interactive simulator where four customizable AI agents compete to find the best staking strategies for eETH. Each Agent can be tuned with parameters like risk tolerance, leverage limits, gas preferences. They independently analyze market data to decide whether to hold, enter, or switch strategies.

How we built it

The app is built with Next.js + TypeScript. Each agent runs its own decision loop that fetches candidate strategies from simulated market data fetched from Merkl API, scores them mathematically using gas cost, leverage caps, and risk adjusted yield. It then calls claude for each day that was simulated for a reasoning model to validate the top pick. This is a necessary step since there are factors that the math itself cannot see, such as external market sentiment, sudden liquidity shifts, or other ecosystem risks that you can only realize intuitively.

Challenges we ran into

Balancing speed and complexity was really difficult, full claude reasoning for every agent each day was actually far too slow, so I implemented a rate limiter to allow only one validation per day per agent. It was also incredibly difficult to design a scoring logic that feels realistic, and often times what would happen is these models would all just end up picking the same strategy within the 5 days simulation, which honestly teaches the user nothing about which strategy was better. I had to find a way to set even harder limits, which is where risk tolerance comes in the picture. It was also difficult getting all the data from all the sources, but I did the best I could with the Merkl API.

Accomplishments that we're proud of

Getting four fully customizable agents running in parallel with distinct behaviors and outputs, and integrating claude in a way that actually influences agent decisions predictably. I'm also really happy I was able to build a working simulation that teaches me about real DeFi tradeoffs like risk vs reward, gas costs, and yield loops.

What we learned

I learned a lot about DeFi lending/borrowing mechanics, and the really long conversations I had with Alex definitely helped me to understand how everything works. On the technical side, I got hands on experience integrating AI reasoning models into an autonomous decision making loop.

What's next for Nilay Kumar - EtherFi Strategy Arena

I would love to expand the project to pull data from more than just Merkl, so that I can get more accurate market data and run better simulations that teach the user far more. I'd also love to add visual performance so that it's easier to understand what is going on behind the scenes.

Built With

• claude
• merkl
• next.js
• typescript