Skip to content
Predictive modeling (churn/subscription) on structured customer dataWe’re using tabular inputs (e.g. features from Recharge/Klaviyo) → MLP is the go-to choice
RBF networks for similarity-based reasoningCascade Correlation architectures for dynamic layer expansionAutoencoders for dimensionality reduction and anomaly detectionRecurrent Networks if you later bring in time-sequenced behavior (e.g. clickstream, email activity logs)
R for data handling and analyticsRecharge, Klaviyo, Shopify APIs for behavioral and transactional dataNeural networks to make predictions (like churn risk or customer LTV)Predict which customers will subscribe or churnSegment users by LTV, location, behaviorIdentify Facebook audiences based on conversion potentialOptimize CAC/CPA by market
Access to real data from a live eCommerce brandThe chance to build, scale, and tune neural networks with production-grade dataAn opportunity to work across data engineering, modeling, and business applicationConstruct a neural engine from scratchUse R and Python in tandemWrite reproducible scripts and train modelsSee how models translate into strategy, ad spend, and revenue
AllowAllow
Business Analytics
🪩 Welcome to the Neural Engine
”but I thought we were working in excel....”
You're stepping into a pivotal moment in modern business: the rise of AI-driven decision-making. At Volcanica Coffee, we're not just watching it happen — we're building it. You are now part of a hands-on, real-world program to analyze customer behavior, forecast outcomes, and help guide a multi-million-dollar brand.
Your mission? Learn how to work with the Neural Engine we’ve developed in R Console on Windows, and eventually take it further.
🧪 What Is the Neural Engine?
Expert Explanation:
Auburn University - Neural Networks Introduction
Dr. Xiaowen Gong ⬇️
ELEC_5240_6240_01_introduction.pdf.pdf2.1 MB
ELEC_5240_6240_01_introduction.pdf.pdf
2.1 MB
ELEC_5240_6240_09_NN_architectures.pdf.pdf5.8 MB
ELEC_5240_6240_09_NN_architectures.pdf.pdf
5.8 MB
“If you read the PDF’s and were able to interpret, we will be using a classic Feedforward Neural Network (FNN), specifically a Multi-Layer Perceptron (MLP)” - Aaron
MLP (Multi-Layer Perceptron) – is simple, effective, and the industry default.
✅ Why MLP Works for You Right Now
🛠️ Later Advanced Exploration
If we ever decide to go further, we might explore:
Easy Explanation:
The Neural Engine is our in-house machine learning pipeline, built on:
You can think of it as a brain built out of data, trained to help us:
You won’t be starting from zero. Aaron has already built and tested foundational models. You’ll build on top of that work — and level it up.
🚀 Why This Matters (and Why It Matters to You)
AI is the defining business advantage of the decade. Companies are racing to build internal data pipelines and modeling tools, but few succeed because it’s hard and takes the right talent.
This internship gives you something rare:
This isn’t school. It’s applied AI.
You’re learning how to:
That’s the skillset top companies and tech-forward brands are competing to hire.
📊 “How Excel & Power BI Fit In”
While the Neural Engine does the modeling, Excel and Power BI are how we visualize, explain, and distribute insights to stakeholders.
Tool Comparison
Tool
Purpose
Excel
Used to manually inspect predictions, validate logic, and run quick pivot-based analysis
Power BI
Used to turn our model outputs into dashboards for our marketing, operations, and executive teams
R Console (Neural Engine)
The actual backend that powers prediction, segmentation, and customer analytics. Quick answer: It spits out CSV’s and models that you will then move into Power Bi
There are no rows in this table
These tools work as one integrated system.
Think: data in, insights out
🚪 Libraries We Use (and Why)
Package
Description
keras3
Builds and trains deep learning models (like predicting churn or LTV)
tensorflow
The core ML engine used to compute neural networks
torch
Alternative deep learning engine used in R for more advanced modeling
reticulate
Bridges R and Python together
tidymodels
Organizes and evaluates models in a reproducible framework
httr
Connects to Recharge, Klaviyo, and other APIs
jsonlite
Converts JSON API data to R-friendly formats
dplyr
Filters, groups, and cleans data efficiently
lubridate
Handles date/time data
There are no rows in this table
❌ What You Will Be Setting Up (Start to Finish)
As part of your learning, none of the infrastructure is pre-set for you. You will go through each setup component with written instructions:
Checklist
Component
Status
R + RTools 42
❌ To be installed and configured by intern
TensorFlow + Python (via venv)
❌ Will be installed and validated in onboarding task
Recharge API Integration
❌ Setup will be done by intern using live key/token
Klaviyo API Integration
❌ Requires newest API revision and will be added from scratch
First Test Model (Churn Example)
❌ Will be built, trained, and evaluated by intern
There are no rows in this table
Every component you set up teaches something critical about how the Neural Engine works.
Ready to get started?
Let’s head over to the first step →
Load content from share.geckoboard.com?
Loading external content may reveal information to 3rd parties. Learn more
Load content from share.geckoboard.com?
Loading external content may reveal information to 3rd parties. Learn more
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.