ForgeFlow
A fictional but realistic manufacturing ERP built on ERPNext
Overview
What is ForgeFlow?
ForgeFlow is the shared reference platform for the Manufacturing AI/ML series. It is a fictional precision engineering manufacturer - FlowForge Industries - that produces industrial components. Built on ERPNext v16, it contains realistic business data across procurement, production, quality, and sales - but no AI. The AI systems are built through the playbooks, one capability at a time.
ForgeFlow is not a demo project. It is a reference platform designed to represent a real client’s business, built on ERPNext v16. It has the data, the workflows, and the operational complexity of a real operation - but none of the AI. The AI systems are built through the Manufacturing AI/ML series, one capability at a time.
Purpose
Why a reference platform?
Standalone playbooks like PageRank or Market Basket teach a technique using a public dataset. You learn the algorithm and see how the numbers work. What they cannot show is how the technique fits into a real system - where the data comes from, how it reaches production, and how it integrates with an existing business.
Series playbooks do something different. Every playbook in the Manufacturing AI/ML series starts from the same business: ForgeFlow. The same master data. The same transactions. Each playbook adds one AI capability on top of it.
This means you can follow the full journey - from raw ERPNext data to a deployed AI system - without re-explaining the business context every time. The reference platform is the shared foundation.
Data Model
Business entities
Raw material suppliers with payment terms and address data - steel, aluminium, consumables, and tooling vendors
Industrial B2B customers across automotive, infrastructure, and OEM segments
8 raw materials (EN8, EN19, EN24, SS304, AL6061, graphite, cutting oil), 15 finished goods (crankshafts, flanges, spindles, couplings)
6 months of procurement data with supplier-specific lead time variance - Amrit (5-7 days) to Vishwakarma (21 days)
Incoming inspection on every GRN. 4 rejected batches from Sudarshan (Feb, Apr) and Paschim (Mar, May) with out-of-spec dimensional and hardness readings
Production orders across 5 finished good families, 6 months, with job cards per operation (turning, milling, grinding, quality check)
6 months of B2B sales with delivery notes - fully linked SO to DN for fulfillment analysis
Architecture
Technical stack
Data access
- Procurement data - purchase orders, receipts, invoices as flat JSON
- Quality inspection data with per-reading results (dimensional, hardness, visual)
- Production data - work orders, job cards, stock entries
- Sales data - orders and delivery notes
- Master data - suppliers, customers, items
Integration
How playbooks use ForgeFlow
Each playbook pulls from ForgeFlow’s data to build and demonstrate one AI capability. The data is the same across all playbooks - what changes is the AI system being built on top of it.
Score every supplier on lead time, quality, and on-time delivery from 6 months of ERP data
Find the 80% on-time miss rate hiding in your ERP work order history and fix it with a lead time calibration model
One supplier holds 66% of your procurement spend - find out before the disruption, not after
Read job card state, detect overdue work orders, find the bottleneck workstation - then let an LLM write the briefing
Quick facts