simply.ai

Understand smart contracts with ease.

Comment

Inspiration

As a novice smart contract developer, I often found myself struggling to understand complex apps. Length, contract inheritance, and technical jargon impeded logic comprehension and vulnerability analysis. This compelled me to create a tool to assist novice developers and non-technical dApp users to better understand smart contracts.

This app enhances the developer experience while empowering users with security-focused insights for informed decision-making.

What it does

simply.ai is an AI-powered smart contract auditing and explanation app. It leverages natural language processing (NLP) algorithms to deliver comprehensive, yet comprehensible explanations and security assessments. Users can upload contract source code (including GitHub repositories) for instantaneous reports: this is particularly beneficial for non-technical dApp/protocol users.

The app supports a myriad of languages including Solidity, Vyper/Python, Rust, Cadence, and non-smart contract languages, proving versatile in even non-blockchain tech stacks. simply.ai has the potential to revolutionize development, enhance user comprehension, and improve risk assessments in product development worldwide.

How we built it

The front-end comprises HTML, CSS (Tailwind), and JavaScript (React); the frameworks greatly streamlined our development process! The web app contains two pages: home and app. These feature the famous "Web3" aesthetic, presenting a stellar UI/UX and paying homage to Web3's love for rounded buttons. Our main page holds briefings regarding app features/mechanics, whereas the app page consists of user input and AI output components.

The back end consists of Python and the GPT-3 large language model (LLM). Python enabled me to develop the core functionality of the back-end, from handling data to processing user requests. I also leveraged OpenAI's ChatGPT API, a pre-trained LLM, to enhance the app's functionality and deliver truly cutting-edge responses.

Challenges we ran into

  • Selecting AI model (GPT3 vs RoBERTa vs BERT)
  • Prompt optimization
  • Pre-trained model (no customizable dataset for training)
  • Solo hacker
  • Novice data scientist

Accomplishments that we're proud of

  • Intersecting AI and blockchain!
  • UI/UX
  • AI research; learning about various LLMs was thrilling!

What we learned

  • Pros and cons of several LLMs
  • Supervised vs unsupervised vs reinforced learning

What's next for simply.ai

  • In-house NLP model
  • Auditing training with web access
  • Chrome extension or Discord/Telegram bot

Built With

Share this project:

Updates