Sourcing Engine [Alpha 1️⃣]

Troy I've tried everything and I can't quite get it the look.

Here is my prompt:
UI / UX of sidebar Source Module
Title: Sources
Three Levels of Tags (make each level of tag a different color, also tags are not exclusive so there can be multiple tags for each source message.)
• Domain: Character, Chemistry, Competency
• Sub-domain: Identity & Purpose, Overcoming Limiters, Teamwork, Problem Solving, Planning & Execution, AE Specific Planning & Execution
• Category: see below.
Direct Links to Sources: Articles / PDFs
Direct excerpts from Sources (one below each source link) 

Categories:
The 7 Habits of Highly Effective People
Assessements
Assessements - ThinkingStyles
Assessements - Listening Styles
Assessements - MBTI (Myers Briggs)
Assessements - DISC
Assessements - Work Genuis
Doden Legacy Statements of Belief
Purpose-Drive Life
Hedgehog
Critical Thinking
Motivation
Emotional Intelligence (self)
Good to Great
Mindset (Growth vs Fixed)
Leadership & Self Deception
Growth Cycle
Dualities
Johari Window
Competency Continuum
Toxic Triangle
I - We - It
Crucial Conversations
Trust Levels
engage!
engage! - discover!
engage! - listen!
engage! - affirm!/challenge!
Emotional Intelligence (others)
Team Member Continuum
collaborate!
delegate!
Communication Levels
Followership
Followership - Appeal
Ideal Team Player (Humble, Hungry, Smart)
Five Dysfunctions of a Team
Death by Meeting (Meeting Management)
The Advantage
Never Split the Difference
Method for Achieving Progress (MAP)
Z-Process
Z-Process - The Six Thinking Hats
Communicating for a Change
Kepner Tregoe Decision Making Grid
Time Management Matrix (Eisenhower)
Porter's Five Forces
Logic Model
Strategic Planning
BrandScript (StoryBrand)
Organizational Essentials
Organizational Essentials - Board GAME
Organizational Life Cycle
Leading Change
Strategic Discovery Process
Growth Platform Strategy
AE Pyramid
AE Pyramid - AE Genome
AE Pyramid - Three Returns
3Cs - Character, Chemistry, Competency
5Ts – Time, Testimony, Talent, Treasure, Temperament
5 Cs: Coach, Connect, Catalyze, Co-create, Conduct

---

Another prompt


Redesign the Source Sidebar in Daryle AI with Domain, Subdomain, Category Tags, Links, and Excerpts

I’m rebuilding the Source Sidebar (right-hand drawer) so that each source message is presented with the following structure and styles:

tsx
Copy

Edit<div className="w-full h-full p-6 overflow-y-auto bg-white shadow-md max-w-md">
  <h2 className="text-xl font-heading text-[#082646] mb-4">Sources</h2>

  {sources.map(source => (
    <div key={source.id} className="mb-8 pb-4 border-b last:border-b-0 last:pb-0">
      {/* Top Tag Row */}
      <div className="flex justify-between items-center mb-2">
        <span className="bg-blue-100 text-blue-800 text-xs px-3 py-1 rounded-full font-medium">
          {source.domain}
        </span>
        <span className="bg-slate-200 text-slate-700 text-xs px-3 py-1 rounded-full font-semibold">
          {source.subdomain}
        </span>
      </div>

      {/* Category / Tags Row */}
      <div className="flex flex-wrap gap-2 mb-3">
        {source.tags.map(tag => (
          <span key={tag} className="bg-[#f6f0e1] text-[#a78940] text-xs px-2 py-1 rounded-full">
            {tag}
          </span>
        ))}
      </div>

      {/* Title + Link */}
      <a
        href={source.url}
        target="_blank"
        rel="noopener noreferrer"
        className="block text-sm font-semibold text-[#082646] hover:underline mb-2"
      >
        {source.title}
      </a>

      {/* Excerpt */}
      {source.excerpt && (
        <p className="text-xs text-gray-600 italic leading-snug">
          {source.excerpt}
        </p>
      )}
    </div>
  ))}
</div>

ElementStyleDomain | bg-blue-100 text-blue-800
Subdomain | bg-slate-200 text-slate-700
Categories | bg-[#f6f0e1] text-[#a78940]
Link | text-[#082646] hover:underline
Excerpt | text-xs italic text-gray-600

ts
Copy

Edit{
  id: "growth-cycle",
  title: "Growth Cycle",
  url: "/documents/growth-cycle.pdf",
  domain: "Character",
  subdomain: "Overcoming Limiters",
  tags: ["Growth Cycle", "Mindset", "Leadership"],
  excerpt: "“Growth is not linear, it cycles through preparation, tension, and renewal.”"
}

Troy

Bug: I'm seeing multiple of the same sources for any given chat or message.

Within a chat, is the (Source: Document, Lines 131-161) actually a thing?

Troy I'm still seeing multiple of the same source. Is that intentional?

No, I think it's the way the data is a. ingested into pinecone or b. on the source drawer logic

will investigate on pinecone

confirmed: it's the way the data is ingested

grabbing the metadata of the title so we won't have duplicates

The metadata might mess up what's retrieved

Troy here is context for these tasks below.

☑️ Only referenced sources for that particular chat

☑️ Update the tag system to include and display Domain, Subdomain, and Category (list here)

☑️ Google Drive vs Pinecone for PDF hosting 

Strangely, without any sources on pinecone, the AI still spits out text without sources. I think it has something to do with the system prompt. Will further investigate

Justin

Confirmed, the system prompt is the problem. It will hallucinate text beyond it's actual knowledge database


next steps: create boundaries in thinking for all prompts

Troy sounds good. Let me know what you find out. 

[TESTING]

I've programmed daryle to become a monkey (for testing purposes)

Daryle will respond "OOH OOH AH AH" when the user inputs something outside the scope of their knowledge database.

Confirmed: Daryle AI is getting his pre-knowledge database from "modeSubdomains," I removed it first to see if it's reading properly from pinecone (and it is)

Even after modifying the base prompts, it still spews out info on growth cycles, which is not referenced anywhere

Will continue to investigate

UPDATE:

Okay, so the issue is the starting prompt:

Defining the boundaries help. But it's push and pull seeing which prompt balances the personality with the boundaries

re ☑️ Google Drive vs Pinecone for PDF hosting 

We're actually storing all our data on pinecone, and just using google drive as the onboarding station

Flow should look like this:

GDrive -> n8n -> Pinecone -> Lovable

[FEATURE: Domains & Subdoins]

- Create edge functions that create domain / subdomain tags based on context

[FEATURE]

I've combined two edge functions into one to reduce latency.

There are two ways to go about the domain, sub domain feature:
1. We manually label them (like this: https://docs.google.com/spreadsheets/d/1ksaNdTJ8m8bZgGvNkJbNLoUF14JKKktmWNdUkW1pPrc/edit?gid=0#gid=0) and tell the system which ones should be labelled which.

Cons: This won't scale if we have a lot of files / sources that we need to work with

2. We use the RAG edge function - when the AI provides an output, it will also generate a tag for each response
Cons: I don't know how accurate it will be everytime (if I talk about character, the tag should be character everytime and not chemistry).

Will continue the test once I upload more documents. But this should fix the problem.

[FEATURE]

Domain, subdomain and categories are now working. Will continue to test

Justin

LET'S GOOOOOOOOOOOOOOOOOOO !!

Added some nice color coding to prettify it

Re email:

Bot refused to answer questions because I only had 2 source documents uploaded. I've reuploaded all of them already.

The bot should answer as close to the user's prompts. The parameters of word search can be adjusted later on to improve the response accuracy (for maximum user satisfaction)

Next steps:

Create n8n workflow that seamlessly ingests source documents / files.

Is there a way to have it just answer if it doesn't have the sources? Troy  

Yes, that will depend on our system prompt.

This was proven already when there was nothing in the knowledge base and the bot was still spewing answers.