JetSetters

LLMs struggle in exotic ecosystems like blockchain. Therefore we enhance `Gemini` with the `Solana Dev Docs`. This new AI agent is the backend for our JetSetters IDE Plugin and the Solanswer AI bot.

Comment

Inspiration

The inspiration for JetSetters x Solanswer came from witnessing the struggles of developers in niche and exotic ecosystems. In these ecosystems, LLMs are often of little help because of missing training data or unique particularities of the environment. An example of such an exotic ecosystem is the Solana blockchain environment. Tools like GitHub Copilot and JetBrains AI struggle to grasp the concepts of blockchain, transactions, and consensus. This makes them give very limited or incorrect answers... until now. Because we wanted to change that!

What did we build

This project consists of three parts.

  1. A workflow to enhance LLMs with documentation and ecosystem-related data via Google Cloud's Retrieval Augmented Generation (RAG) system. As an example, we enhance Gemini 1.5 Pro with the Solana Developer Documentation.
  2. The JetSetters Plugin for JetBrains IDEs (see installation guide on GitHub)
  3. The Solanswer Website (https://solanswer.vercel.app/) as a web version of our AI agent

How we built it

We started by compiling the publicly available Solana Developer Documentation into a vector database on Google Cloud. With this database, we used Retrieval Augmented Generation (RAG) to add it to an existing LLM (Gemini 1.5 Pro). After wrapping our newly enhanced LLM in an API we started building two frontend products: The JetSetters Plugin and the Solanswer Website. Both hock into our Google Cloud backend.

The JetSetters Plugin provides developers with instant AI-driven documentation and code insights directly within the JetBrains IDEs.

The Solanswer Website (https://solanswer.vercel.app/) provides a web version of our AI agent. It is capable of answering in-depth questions regarding the Solana Documentation.

Challenges we ran into

One of the major challenges we faced was the integration of the RAG system with our plugin. Ensuring that the chat bot could accurately retrieve and present relevant documentation in response to specific queries required extensive testing and debugging. Additionally, we encountered hurdles in maintaining the responsiveness of the output field while dealing with large volumes of data.

Accomplishments that we're proud of

We successfully developed a functional prototype of our AI Agent for real-time interaction with the Solana documentation. For many questions regarding the Solana Docs our AI provides more in-depth answers than GitHub Copilot and JetBrains AI. This proves that our product fills a viable market niche where it can help make the lives of developers easier.

Our achievements also include creating a user-friendly interface and receiving positive feedback during the initial testing phases.

What we learned

This project taught us the importance of collaborative problem-solving and the value of user feedback in refining our application. We also gained insights into the intricacies of plugin development and the potential of AI to enhance developer productivity.

What's next for JetSetters x Solanswer

Moving forward, we plan to expand our AI Agent's capabilities by incorporating features such as automated code reviews, security checks, and enhanced support for additional programming environments and frameworks. We aim to continuously improve our AI model's accuracy and expand the range of resources it can access, ultimately creating a comprehensive tool that significantly enhances the developer experience in exotic ecosystems. Furthermore, a direct implementation of Solanswer into the Solana Docs website could make it even more accessible for developers.

Built With

Share this project:

Updates