Path B: Building AI Systems
Last updated: 2026-08-04
Overview
- Audience: developers, data and BI people working in e-commerce
- Prerequisites: some Python (or willingness to learn as you go — the AI will help you write it)
- Time: 1 hour a day, 4–8 weeks to work through it
- Output: a deployable AI tool
Build AI-driven e-commerce tools and systems, from scripts to production applications
flowchart LR
B1["B1 Data Collection\nand Processing"] --> B2["B2 Prediction Models\nand Decisions"]
B2 --> B3["B3 RAG\nKnowledge Base"]
B3 --> B4["B4 AI Agents\nand Automation"]
B4 --> B5["B5 Local Model\nDeployment & Tuning"]
Module navigation
| Module | Topic | Difficulty | Time | What it covers |
|---|---|---|---|---|
| B1. Data Collection & Processing | Data pipeline | Beginner | 4–6 h | From Amazon reports to a cleaned analysis dataset |
| B2. Prediction Models & Decisions | Predictive modelling | Intermediate | 6–8 h | Sales forecasting to support restock decisions |
| B3. RAG Knowledge Base | Knowledge base | Intermediate | 6–8 h | An AI Q&A system over your internal documents |
| B4. AI Agents & Workflow Automation | Agents | Advanced | 8–10 h | Executing multi-step operational tasks automatically |
| B5. Local Model Deployment & Fine-Tuning | Model deployment | Advanced | 4–6 h | Run an LLM locally, keep the data in-house |
| B6. MCP Integration & Agentic Workflows | MCP / agentic | Advanced | 2–3 weeks | Connect Amazon Ads / Shopify over MCP, run operations by conversation |
| B7. Review Analysis System | NLP / topic modelling | Intermediate | 2 weeks | BERTopic topic modelling + sentiment analysis + LLM insights |
| B8. E-Commerce Dashboard | Streamlit / Plotly | Intermediate | 1–2 weeks | Multi-platform operations dashboard + AI anomaly detection |
| B9. AI Product Image/Video Generation | ComfyUI / GPT Image 2 / FLUX.2 | Advanced | 2–3 weeks | Batch product-image pipeline + video generation |
Progress tracking
[ ] B1. Data: write a script that merges several Amazon reports and produces a summary
[ ] B2. Forecasting: run a 90-day sales forecast for a real SKU with Prophet
[ ] B3. RAG: stand up a RAG system that answers product questions
[ ] B4. Agents: deploy an agent that monitors operations automatically
[ ] B5. Deployment: run an LLM locally with Ollama and complete one e-commerce task (optional)
[ ] B6. MCP: configure the Amazon Ads MCP and manage ads through a conversation with Claude
[ ] B7. NLP: topic-model 1000+ reviews with BERTopic
[ ] B8. Dashboard: build a Streamlit operations dashboard with 4+ modules
[ ] B9. Images: generate a full AI image set for one product and pass Amazon's compliance check
Related resource: Technical Implementation Guidelines architecture patterns, performance benchmarks, and a security/compliance checklist.
Path B is done when: you have completed at least 3 of B1–B4 — at that point you can build AI e-commerce tools. B5–B9 are there when you need them: B5 keeps data in-house, B6 turns operations into a conversation, and B7–B9 are each a finished system of their own.