Developer Support Bot Trained On Decentraland Documents

by 0xce6bf74d6b9d998e5f44364899dabe3cb3ba8497

Should the following $5,000 grant in the Documentation category be approved?

Abstract

We propose to build a Discord bot for the Decentraland developer community that will use the latest technology around LLM’s (Large Language Models) to import Decentraland's Github repos, documentation, Slack and Discord Q&A’s to provide an interface for Decentraland developers and users to ask questions in natural language.

Our goal is to provide another member to Decentraland's DevRel team, reducing support hours on Discord and Slack and a path of triage for commonly asked questions (making the team and users get answers quicker).

Grant size

5,000 USD

Project duration

1 month

Beneficiary address

0xc5F1C802F56e34B05e9aAfBFE0302c6595405035

Email address

ali@trytali.com

Description

The core of technical Web3 projects depends on a healthy ecosystem of developers using the protocols. Developers learn how to use the various projects through inspecting developer docs and Github repositories. As expected, documentation doesn’t cover everything and a team of people are needed to provide “front line” support to those building/using the protocol.

Enter Developer Relations, a team tasked with being the first human interface with protocols who provide 2 critical roles for the project: building developer communities & tracking product-market fit.

The issue that often arises is a lack of human capital to be able to deal with the number of requests & questions that populate a DevRel team's day.

We propose to provide a solution that will enable the Decentraland DevRel team to be able to focus on what they do best: solving unique customer problems and building trust in the protocol. By providing a bot that can answer previously asked questions, it will enable DevRel to become more effective and in turn provide higher quality support to end users.

Roadmap and milestones

We will use the funds to cover our development time (four weeks) and hosting for the bot for 6 months.

Project Plan: Pre-Implementation: We’d like to work with the DevRel team to understand their needs, sticking points, and time sinks to see where the bot can be of most assistance. We’ll use these insights to see which data sources need to be implemented.

Milestone 1: Implementation of Data Sources [2 weeks]: Once discovery has been done, we’ll spend the time implementing the 3-4 most high-value data sources to Tali. We will leverage Llama Index to create indexes and a vector database to enable embeddings based search on queries. Optimization around how files are loaded, index creation, and integrating various data sources will need to be tested.

End state: Have 3-4 data sources that can be queried on. Expected bugs: hallucinations, mismatch on query/documents.

Milestone 2: Testing + Optimization [2 weeks]: Arguably the most important part of the process. The focus will be to discard hallucinations and optimize prompt engineering so only relevant answers appear, ideally with source material & links for the user to follow up on. The purpose of this will be to make the bot production ready.

End state: Production-ready bot ready (once signed off on by DevRel team) to be deployed to Decentraland Discord and or Slack.

This proposal is summarized due to technical limitations. To view it complete and vote on it, visit the DCL DAO Governance dApp