Evergreen Capital

Fully automated on-chain AI predictive market hedge fund

Comment

Inspiration

We wanted to build something that mixes AI, trading, and Solana in a fun and automated way. Prediction markets are full of opportunities, so we built a system that can actually catch them without needing people online 24/7.

What it does

Evergreen lets anyone deposit SOL into a shared pool. We scrape Polymarket markets and web data, store everything in Snowflake, and send it to five Gemini agents with different personalities. They research, debate, and vote on whether a trade is a buy or no buy. If they vote yes, the system places the trade and later sends winnings back to depositors based on their share of the pool.

How we built it

We built a pipeline that scrapes Polymarket and online sources. All data goes into Snowflake where we clean it and prepare it for the agents. Then we run a multi agent debate using Gemini and feed the final decision into a Solana execution module that handles deposits, trades, and payouts.

Challenges we ran into

Getting Snowflake to connect was way harder than expected. Making agents debate in a realistic way also took tuning (so many prompt changes). Onchain execution and timing around Polymarket markets created a lot of edge cases.

Accomplishments that we're proud of

We built a full AI trading loop that goes from data to debate to automatic execution. The team also got a clean Snowflake setup running at top speed. The multi agent system produced noticeably better decisions than a single model. I dont think a lot of teams would be brave enough to try something this large in a 24 hour hackathon.

What we learned

Good data matters more than anything. Multi agent setups are powerful but need structure. Solana is great for fast, cheap execution. Prediction markets reward consistency.

What's next for Evergreen Capital

Add more data sources, launch a public dashboard, expand to more prediction platforms, and eventually open Evergreen to anyone who wants passive AI powered returns on their SOL.

Built With

Share this project:

Updates