We propose to develop a private Discord bot or Docs widget (allowing users to “talk to your docs”) for Hxro Network, leveraging the capabilities of the latest Large Language Models (LLMs). This bot or widget will integrate Hxro’s Github repositories, documentation, related articles, and Discord Q&A history, offering users a platform to ask questions in a natural language. This solution will bolster Developer Relations, Customer Support, and Community Support experiences within the Hxro and broader Solana ecosystem.
Having been approved for a grant from the Solana Foundation, we are committed to contributing to the development and growth of the Solana ecosystem. Our primary goal is to collaborate with Hxro, an integral player within the ecosystem, to further improve their DevRel and Community Support teams. We aim to develop a solution that will reduce support hours on Discord, provide prompt responses to frequently asked questions, and enhance user interaction and satisfaction. This commitment aligns with our broader vision of fostering an efficient, robust, and productive Solana ecosystem, and ensuring that both Hxro’s teams and users benefit from streamlined and effective support.
Background
The success of technical Web3 projects hinges on a strong ecosystem of developers using the protocols. Developers familiarize themselves with various projects by reviewing developer documentation and Github repositories. Yet, documentation can’t cover everything, requiring a team of individuals to offer “front line” support to those developing or using the protocol.
Developer Relations and Community Support teams are entrusted with being the first human interface with protocols. They play two crucial roles in the project: cultivating developer communities and monitoring product-market fit.
A common challenge is the insufficient human resources to handle the numerous requests and questions that teams receive daily. Our solution aims to empower Hxro’s support teams to focus on their core strengths: resolving unique customer issues, building trust in the protocol, and fostering a vibrant and engaged community. By introducing a bot or widget capable of answering previously asked questions, support teams can become more efficient and provide superior assistance to end-users.
Team
Ali Agha is a technologist and entrepreneur focused on decentralized solutions. In his previous venture, Olypsis Technologies, Ali offered Web3 consulting services to countless startups and big corporations like IBM and Thomson Reuters. Ali started his journey in the blockchain space in 2015 when he discovered bitcoin. Since then, he has dedicated his career to creating a fairer and more equitable world through the power of decentralization.
Github:OlypsisAli · GitHub
Twitter: @iamAliAgha
Tenzin Rose is an entrepreneur and full-stack developer with a background in enterprise sales. He’s worked with global startups and enterprises, helping them successfully deploy projects and generate revenue. His current interests lie in web development and deciphering the complex equations in ZKP and ML.
Github: @niznet89
Twitter: https://twitter.com/tenzin_rose
Project Plan
Pre-Implementation: Collaborate with the DevRel, Customer Support, and Community Support teams to understand their needs, identify pain points, and determine where the bot or widget can be most useful. These insights will guide the choice of data sources for implementation.
Milestone 1 - Implementation of Data Sources [3 weeks]: After the discovery phase, we will integrate 3-4 high-value data sources into the bot or widget. We will use the Llama Index to set up indices and a vector database to enable embeddings-based search on queries. This milestone will involve refining file loading, index creation, and data source integration.
End state: Have 3-4 data sources that can be queried.
Expected bugs: hallucinations, mismatches on query/documents.
Milestone 2 - Testing and Optimization + Deploy to Production [3 weeks]: This critical phase will focus on eliminating hallucinations and refining prompt engineering to ensure only relevant answers are provided, with source material and links for users to follow up on. The goal is to make the bot or widget production-ready.
End state: A production-ready bot or widget approved by the Customer / Community Support teams, ready for deployment on Hxro’s Discord channel or Docs page.
Technical Details:
The proposed Discord bot or Docs widget will be developed using the latest Large Language Models (LLMs) and other state-of-the-art technologies. Here’s an outline of the technical aspects involved in the project:
Data Collection and Indexing: This phase involves gathering data from Aleph Zero’s GitHub repositories, official documentation, relevant articles, and past Q&A from Discord. We will use the Llama Index to establish indices and a vector database to facilitate embeddings-based search on queries.
Bot Development: We plan to use Python for the bot’s back-end development. This involves the integration of LLMs like OpenAI for understanding and generating responses to user queries. In the front-end, we will ensure seamless integration with Discord or as a widget for the documentation page.
Optimization: This phase includes refining prompt engineering and eliminating any hallucinations or mismatches. We’ll be utilizing tools like Deeplake and Cohere for enhancing performance and achieving optimized results.
Deployment and Maintenance: The bot will be hosted on Digital Ocean, ensuring high availability and performance. Regular maintenance, updates, and improvements will be part of our ongoing commitment.
Funding Requirements:
We request a $5,000 USD grant to build and maintain the bot. The budget allocation includes:
Infrastructure costs: 12 month provision (depending on usage)
OpenAI token usage
Digital Ocean hosting
Supabase, Deeplake and Cohere usage
Potential for contract work from Data Science/ML experts if needed.