Stoop Dynamics
EVOLVE Dashboard

EVOLVE AI Assistant

Production AI system for the #1 e-commerce community on Skool

Built for Shaun Eng & Spencer Pawliw1000+ Active Members | Teaching $100k/day Revenue

Impact & Results

24/7
Availability
Instant expert-level support anytime
100%
Citation Rate
Every claim backed by sources
110K+
Context Size
Messages + course docs searchable
~2m
Response Time
Average pipeline completion

The Challenge

EVOLVE is the #1 Skool community teaching advanced Shopify sellers to scale to $100k/day. With 1000+ active members, they needed:

  • 24/7 expert support that couldn't be delivered manually at scale
  • 110K+ Discord messages + Skool course materials searchable for accurate, cited answers
  • Zero tolerance for errors - wrong advice at $100k/day scale means major financial risk

The Solution

Production AI system with three components:

1. Discord Bot ("Chad")

  • RAG with dual-retrieval (FTS + semantic)
  • Multi-stage verification + hallucination detection
  • Question decomposition for complex queries
  • Forced citations for every claim
  • Smart retry with gap-based refinement

2. Analytics Dashboard

  • Real-time metrics and engagement analytics
  • Semantic heatmap of conversation topics
  • Knowledge base management with AI search
  • Channel configuration and settings

3. Content Agent

  • Discord search for community insights
  • X/Twitter trend analysis
  • Semantic content mapping
  • Memory for iterative ideation

Technical Architecture

Multi-stage RAG pipeline prioritizing accuracy:

1

Question Analysis

Breaks complex questions into 1-3 subquestions. Extracts entities and intent.

2

Dual Retrieval

FTS for precision + semantic search for coverage. Top 10 from each, deduplicated.

3

LLM Selection

Filters for truly relevant spans. Generates gap queries if coverage is low.

4

Document Integration

Adds full docs (top 3) + highlights (up to 10 excerpts) from knowledge base.

5

Generate & Cite

Creates answers with forced inline citations [C1], [D2] for every claim.

6

Verify & Retry

Confidence scoring + hallucination check. Auto-retry if confidence < 0.8.

Key Innovations

Question Decomposition

Breaks complex questions into subquestions for separate retrieval. Dramatically improves completeness.

Gap-Based Retry

Auto-generates targeted queries when coverage is low. Self-correcting without manual intervention.

FTS + Semantic Hybrid

FTS for precision on keywords, semantic for broad coverage. Best of both worlds.

LLM Selection Filter

Extracts only truly relevant spans instead of relying on similarity scores alone.

Semantic Heatmap

Visualize conversation topics and trends

Semantic Heatmap

Analyzes all Discord messages to create interactive heatmaps showing topic clusters and how conversations evolve over time.

Topic Clusters

Auto-groups similar conversations. See ad strategies vs product selection vs scaling tactics at a glance.

Trend Analysis

Weekly/monthly views show shifting interests. Spot emerging topics early.

Content Strategy

Shaun & Spencer use this to guide social media and course content creation.

Interactive

Click clusters to see actual messages with full context.

Technical Implementation: UMAP + HDBSCAN clustering on embeddings. Generates static HTML stored in Supabase.

Technology Stack

Backend

Python 3.14discord.pyasyncioOpenRouter APIOpenAI API

Frontend

Next.js 15React 18TypeScriptTailwind CSSshadcn/ui

Database

SupabasePostgreSQLpgvectorFull-Text Search

Infrastructure

Vercel (Dashboard)Fly.io (Bot)N8N (Dashboard Agent)Supabase Storage

Ready to build your AI solution?

Get in touch to discuss how we can help your business.

Get in touch