The problem TrailBlazers solves

ETH24 Project Overview

At ETH24, we tackled challenging problem statements from Huddle01 & LIT. Our key achievements include:

LLM Integration

• Challenge: Integrate Anthropic-based text models into Huddle01's OpenAI-based infrastructure, initially supporting only audio models.
• Solutions:
  • Navigated and modified the extensive codebase.
  • Created new abstract classes and extended existing model options to support text communication.
  • Enhanced infrastructure to handle dual input types seamlessly.

Real-Time Trading Bot

• Development: Built a sophisticated real-time trading bot as part of Huddle01's AI track.
• Architecture:
  • Utilized Huddle API to connect a master LLM with specialized industry expert LLMs.
  • Based on user input, the master LLM coordinates with experts to gather relevant context.
  • Updates parameters for the algorithmic trading bot to execute trades effectively.
• Impact: Delivered one of the toughest tracks characterized by Huddle01, showcasing our expertise in AI-driven trading solutions.

LIT

• Innovative Gas Management Solution: GasBank is a cutting-edge platform built on the Lit Protocol that functions as a "gas bank" for cryptocurrency wallets, enabling users to perform on-chain transactions without the necessity of holding native cryptocurrencies to cover gas fees.

• Fiat Currency Integration: By allowing users to pay gas fees using INR or other fiat currencies through reliable payment gateways such as Razorpay and Juspay, GasBank bridges the gap between traditional financial systems and the blockchain ecosystem.

• Secure Payment Verification: Utilizing the advanced security features of the Lit Protocol, GasBank ensures that all fiat transactions are securely verified and that sensitive payment data is managed with utmost confidentiality and integrity.

• Automated Transaction Submission: Once a user's payment is successfully verified, GasBank s

Challenges we ran into

With Huddle, there was a lot of hustle. This is because the repository given initially did not even have a single runnable example and I had to navigate through all the API documentation and add changes to internal library functions (like the _lock inside the call_handlers function which was internally used). Further, this task was ginormous as it involved providing new infra for anthropic based models. These models are text based but the existing openai framework used audio based models, so I had to rewrite a lot of classes and extend lot of existing code in the given pipeline to get things working. I was working in the debugger mode (using pdb) and making incremental changes to few classes. I adapted the openai realtime codebase to anthropic and made incremental changes to finish it. I also discussed ideas with partners from Huddle and figured out the pipeline for trading bot and how to simulate it. Finally, over the course of night, I was able to implement the master LLM and specialised agent to complete the trading bot.
For LIT protocol, the challange was the ideation part of how can we use this innovation in efficient gasless transactions. We wanted to build something that imparts significant value to people both new and experienced in this crypto space. The problem of gas has been a huge blocker in onboarding newcomers to web3 and use the benefits of web3.