LaraLamma

Appearance

LaraLamma ​

LaraLamma is based on Laravel and is a great foundation for any LLM centric application especially a RAG system.

The code is here https://github.com/LlmLaraHub/laralamma

INFO

A RAG system (Retrieval augmented generation system (RAG - an architectural approach that can improve the efficacy of large language model (LLM) applications

It really can be more than a search mechanism for your business.

  • Want to surface the most common customer feedback requests just send them to your Collection.
  • Want to centralize emails and route them to the correct support team just have a Source check that email box.
  • Want to Chat with your data on another site, just point the included chat widget to the Output on that collection.
  • Have a shared marketing inbox, have all those messages summarized and sent to the right people

INFO

LLMs are large language models

Three Key Concepts ​

Source ​

This is how you get data into the system. There are many ways to do this.

  • Upload files (PDF, PPT, Text etc)
  • Send in data as emails (the system can query an email box or you can forward to it)
  • Send in data via a webhooks (Create for GitHub data)
  • WebPage Source - here you can enter a list of URLs for the system to get the data from

Trasnformers ​

The system will already vectorize (make sure data easier for the Language Models to search) and tag but you can add more Prompt based trasnformers to alter your data. A great example can be seen here https://youtu.be/KM7AyRHx0jQ?si=vnYG0PE8I8OCVEQa as the user can write a "Prompt" to take complex JSON data from GitHub and turn it into a simple summary.

TIP

Prompting enpowers the user to write plain language instructions for the computer to do.

Output ​

After you have all your data in the system there are so many ways you can get to it and use it.

  • Log into the system and Chat with your collection
  • Have it send you daily, weekly summary emails
  • Post to a website the updates
  • React to a customer support ticket by replying to a message based on all of it's data on the related domain

About Your Data ​

Below is what the system does with the data you add to it to make it more LLM friendly.

Collections ​

Collections make up the “Chattable” grouping of Documents. You make a collection and add documents to it.

Since this system uses Laravel JetStream https://jetstream.laravel.com/introduction.html we can share those collections with others on the team.

Documents ​

This is the foundation to the searchable content. It can come from PDFs, Websites, Power Points, Text and more.

When a document is added the system will do the following.

  • Break it into Chunks (small fragments) that overlap with the sibling before and after it.
  • Do vector embedding on it so we can query it later

Document Chunks ​

These are related to documents. They have the chunk of text from the document. A page number and a sort number. This always has the vector field we do all the searching on.