Forge 6 Builder Handbook · Industrial AI

Forge 6 Handbook.

For builders of nmg.labs Forge Sprint 06 · Industrial AI · Powering the Shopfloor · NMG Gurugram · date to be announced. Read this before the day; pre-read the use-cases and set up your stack.

Get the Forge 6 date drop →

On this page
Who it's for The nine use-cases Stack & setup The day Scoring Demo checklist Rules

Who it's for

Forge 6 is for builders who want to make AI that does real industrial work: catch a defect on a line, draft a quality document in seconds, predict which supplier is about to slip, tell a plant what a part should cost. Final-year and pre-final-year CS / IT / data / AI students and especially mechanical / production / industrial / mechatronics students whose domain knowledge is a superpower here. Solo or pods of up to 3. Mixed pods (an ML person, a domain person and a builder) tend to win industrial problems.

The nine use-cases

These are the AI use-cases the auto and manufacturing industry is actively prioritising for its supplier base. Pick one. ⭐ flagship tracks are most demo-able in a day; 🏆 is the Grand Challenge.

Track	Mission	KPI
A · Eagle-Eye QC ⭐	Camera that spots and classifies defects on a part, live	Defect-rate down
B · Quality Co-Pilot ⭐	AI that drafts DFMEA / APQP / PPAP docs + test cases	Time-to-market down
C · Should-Cost Engine ⭐	Estimate what a part should cost, to localise sourcing	Localisation up
D · Guardian Vision	Catch a missing helmet, restricted zone or near-miss	Incidents down
E · Flow State	Optimise stock across the network, re-route logistics	Working capital up
F · Early Warning	Predict which supplier is about to drift	Handling time down
G · Green Ledger	Turn carbon rules (CCTS/PAT) into an emissions dashboard	Future compliance up
Grand · First-Time-Right 🏆	Predict how a component behaves before it's built or an agentic shopfloor co-worker	First-time-right up

You don't need to be a factory expert. Pre-read your chosen track; the specific problem variant and starter dataset drop at kickoff.

Stack & setup

Anchors

An AI coding co-pilot (Claude Code, GitHub Copilot, ChatGPT, Gemini or an open LLM) and the brain for the GenAI co-pilot + agentic tracks.
A vision model (YOLO-class detector / lightweight CNN) for QC and safety-vision, runs on a laptop + webcam.
Python data stack (pandas, scikit-learn, a forecasting lib) for should-costing, inventory, supplier-sensing and carbon analytics.
A RAG setup (vector store + your LLM) for the Quality Co-Pilot drafting real documents.

Encouraged shelf

Hugging Face models · Ultralytics/YOLO · OpenCV · LangChain / LlamaIndex or an agent framework · Hermes-style orchestration + OpenClaw · a Streamlit / Gradio front-end for the demo · any open optimisation / solver library.

Before the day: install your stack, pull a base vision model and run a hello-world inference. Bring your own laptop (a webcam helps for the vision tracks).

The day (Forge Clock)

09:00	Check-in · stack sanity check · pod formation
09:45	Opening · the brief + datasets drop
10:15	Forge Clock starts · scope & sketch (design your demo first)
11:00	Build Block 1 · crude end-to-end thing running
14:00	Build Block 2 · improve, tie to KPI, make it legible
16:30	Freeze + record your 30-60s showcase clip
17:00	Hard stop, commit your repo · demos (3 min + 2 min Q&A)
18:30	Winners · showcase notes · NMG fast-track

Scoring (100 points)

Working prototype (runs / demos live), 20
Shopfloor impact (KPI moved), 20
Use of AI & technical depth, 20
Real-world fit & data rigor, 15
Productionizable for a supplier, 15
Demo & storytelling, 10

Bonus +5: deployable on a real line (edge-ready / commodity HW / integration path). Judges: NMG senior engineers + invited automotive and industry guests. Demos are recorded for the vendor showcase reel.

What "done" looks like

It runs live (with a backup recording).
You can state your track's KPI and show it moving.
You have a one-line "without AI vs with AI" number.
A non-engineer (plant manager) could follow your 3-minute demo.
Your repo is committed and your 30-60s showcase clip is recorded.

Rules

Solo or pods of up to 3. Lock your pod before the clock starts.
All building happens on the clock. Pre-reading and stack setup are expected; pre-building the solution is not.
Use any tools/models you like (open or commercial). Cite anything substantial you didn't write.
It must run. Live demo or it doesn't count; have a recorded backup.
Tie it to the KPI. Provided datasets are the baseline; public datasets allowed; no proprietary/confidential data.
Be decent. Respectful, safe, your own work. Mentors guide, they don't build for you.

Questions? Forge Captain Ayush Gupta · labs@nmgdigital.com.

Get the Forge 6 date drop →