spike: faq-ai

[joepegler]

• Investigate solutions for ai FAQ
• Speak with Graeme
• Claude?
• Embed at bottom of "batch txs" vocs page

[Graeme-Code]

Yo! @B-Venkatesh7210 please add some time to my calendar. I can save you a bunch of time setting this up.

[Graeme-Code]

Hey @B-Venkatesh7210 @joepegler. Thought a bit more about this and want to add some actionable outcomes from this spike.

1. Create a Replit Template for Nexus and Smart Sessions.
https://replit.com/~
Replit templates are pre-configured starter projects provided by Replit, an online IDE that enables developers to write, run, and collaborate on code directly in the browser. These templates come with predefined environments, dependencies, and example code tailored to specific programming languages, frameworks, or project types.

Onboarding with Replit Templates

Using Replit templates can streamline the onboarding process for new developers by offering a ready-made setup aligned with our new technology stack. Junior developers can quickly start working on projects without the hassle of initial configuration, ensuring they have the correct dependencies and project structure from day one. This approach accelerates learning, reduces setup time, and allows new team members to focus on contributing to the project immediately. Additionally, templates can be customized and shared within the team, promoting consistency and best practices across our development workflow.

Hypothesis: If we have replit templates for Nexus and Smart Sessions, we will reduce the time required for developers to start using our stack.

2. Create a n8n vector store powered AI agent to answer questions about Nexus, Smart Sessions and Infra.

Check out this tutorial on how to use a no - code tool ensure an AI references data via a vector store to answer questions.

https://www.youtube.com/watch?v=nE7XATKx6iQ

This will allow use to see how useful or how good an AI agent can be answering users questions when it is fed with context specific to our product.

3. Identify data sources for AI flywheel.

If an AI is going to be leveraged understanding what data we have now, how we can get more and how we can organize it will give outsized returns. Item two on this list requires uploading of data that we know (dev docs, repos, error codes), however there is a wealth of information around debugging issues that we gain in discord, telegram and on github issues. If this data can be loaded to a vector store not only could our AI answer questions related to our code, it will have reference points of how issues have been debugged in the past. Over time this can save you a ton of time as well as our customers.

Grooming:

1. likely makes sense that Replit Templates is a seperate issue and perhaps better suited for @himanshugarg06 given she is writing the tutorials.

2. N8n vector store test would need to be its own task.

3. Mapping our customer technical assistence data would also be its own issue. I see having three parts:

• Where do we have debugging data currently
• How could we streamline the collection of this data so it is easy to work with.
• How could we automate the inclusion of it in a vector DB.