Inspiration
The inspiration from the project comes from various automation tasks happening around right now. With the rise of various meme-coins and AI agents people feel they could do it better. Through this chain, and various contests organized on top of it we want to give the power to user to put their money and even the stakes with algorithms and AI agents in a fair, transparent, and competitive environment.
What it does
CasinoNet is a virtual chain on Aurora that will introduce AI agents to act on your behalf. Make calculated moves, define your own risks and combine your skills with the agent tech. Powered by Near-AI and Aurora Virtual Cloud the chan will introduce various competitions verifiable on-chain, and allow users to participate using their own agents.
The first of many games is a Blackjack game - A dealer agent with off-chain randomness and card dealing, it takes a user agent name and allows user to play with the BlackJack dealer. Users can create and name their own agents to compete in a 1v1 match against the dealer. Currently, the game is designed as a win-or-lose format with no monetary stakes.
How we built it
Mainly made use of the Near-ai agents directory and docs, with the llama-v3p1-70b-instruct model.
Challenges we ran into
Building CasinoNet’s Blackjack Agent required multiple iterations to fine-tune the AI and ensure fair, transparent gameplay on-chain. Throughout development, we tackled challenges related to AI decision-making, randomness, and smart contract integration.
Agent Development
Our AI dealer follows standard Blackjack rules, but early versions suffered from hallucinations, making unrealistic moves or misinterpreting the game state. To fix this, we refined its decision-making through a mix of reinforcement learning, rule-based constraints, and extensive testing. There were cases where the Agent frogot the cards and hence we had to form the full gamestate in a file.
User agents were also optimized to allow strategy customization, giving players control over factors like risk tolerance and betting style.
Off-Chain Randomness & Fairness
Since blockchain environments struggle with native randomness, we're following off-chain randomness for fair and unpredictable card dealing.
What's next for Casinonet
CasinoNet is just getting started. Our next steps focus on improving transparency, gameplay mechanics, and expanding AI-driven casino games.
- Enhancing Smart Contract Transparency
* Improve contract interactions to ensure greater verifiability.
* Implement preimage deck verification, allowing players to confirm that the deck was not manipulated before the game started.
- Introducing Betting & Casino Economy
* Add staking mechanics and introduce a casino chip system for gameplay.
* Enable on-chain betting and payouts, making AI-driven casino gaming more engaging.
- Expanding Game Selection
* Move beyond Blackjack and introduce AI-driven poker, roulette, and other casino games.
* Allow players to fine-tune AI strategies across multiple games.
- AI Strategy Optimization & Tournaments
* Implement leaderboards and AI competitions to encourage strategic gameplay.
* Utilize Swan Chain for AI-driven insights into player strategies.
Built With
- nearai
- next.js
- python
Log in or sign up for Devpost to join the conversation.