I'll help organize this information under the hackathon-style headers you provided.

Inspiration

The blockchain industry faces significant challenges with regulatory compliance, as developers must navigate complex legal requirements when creating smart contracts. We recognized that manually checking regulations and implementing them in code is time-consuming and error-prone. This inspired us to create a solution that bridges the gap between legal documentation and smart contract development.

What it does

funtek is a compliance solution that uses a two-LLM approach to transform legal requirements into executable smart contracts:

1. The first LLM processes legal documents, interpreting complex regulatory text into clear, actionable items
2. The second LLM, specialized in Solidity, converts these actionable items into secure, compliant smart contract code

This automated process ensures regulatory compliance while significantly speeding up smart contract development.

How we are building it

We developed our solution using:

• GPT-2 for text generation capabilities
• Facebook-BART-base model for text summarization
• Python and Hugging Face libraries for implementation
• Custom dataset creation using GitHub API to gather Solidity code snippets and their corresponding documentation
• Integration of legal text interpretation and Solidity code generation models

Challenges we ran into

Creating appropriate datasets for training our models was a major challenge. We needed:

• Input-output pairs matching legal text to interpretations for the first model
• Task descriptions paired with Solidity code for the second model We solved this by developing a custom Python script using the GitHub API to scrape repository README files and match them with corresponding Solidity code snippets.

What we learned

• The complexity of combining legal interpretation with code generation
• Techniques for dataset creation and curation
• Implementation of multiple LLM models in a single solution
• The importance of maintaining accuracy when translating between legal and technical domains

What's next for funtek

Our roadmap includes:

• One week of comprehensive dataset preparation and cleaning
• Two weeks of model training and performance evaluation
• Two weeks dedicated to developing a website with Reinforcement Learning with Human Feedback (RLHF) capabilities
• Continued refinement of the model's accuracy in both legal interpretation and code generation
• Expanding the range of supported regulatory frameworks and smart contract patterns

Built With

• google-ai-studio
• hugging-face