Tutorials

The tutorials are step-by-step guides that assume you are new to TrainCraft. Each one introduces a specific system type or capability and builds on the previous ones.

You don't have to read them in order — jump to whichever system type is most relevant to your work.

---

Learning path

graph LR
    T1["1 · First Dataset<br/>(CNT + EMT + MD)"]
    T2["2 · Molecules<br/>(SMILES + RDKit)"]
    T3["3 · Surfaces<br/>(adsorbate + MC)"]
    T4["4 · Crystals<br/>(defects)"]
    T5["5 · Slabs<br/>(strain + transforms)"]
    T6["6 · 2D Materials<br/>(graphene, MoS₂)"]
    T7["7 · MACE-MP0<br/>(foundation model)"]
    T8["8 · Selection Funnel<br/>(diversity, budget)"]
    T9["9 · Interop<br/>(pymatgen, RDKit)"]
    T10["10 · Training<br/>(fine-tune MACE)"]
    T11["11 · AI Agent<br/>(Pi + Gemma + visualise)"]

    T1 --> T2
    T1 --> T4
    T2 --> T3
    T4 --> T5
    T4 --> T6
    T1 --> T7
    T1 --> T8
    T3 --> T9
    T7 --> T10
    T8 --> T10
    T1 --> T11

---

Quick reference

Tutorial	System type	Extra deps	Pixi env
1 · First Dataset	Carbon nanotube	None	default
2 · Molecules	Small molecules, SMILES	RDKit	science
3 · Surfaces	Adsorbate + MC sampler	(RDKit optional)	default / science
4 · Crystals	Bulk + defects	None	default
5 · Slabs & Strain	Slab + transforms	None	default
6 · 2D Materials	Graphene, hBN, MoS₂	None	default
7 · MACE-MP0	Any periodic system	torch, mace-torch	mace
8 · Selection Funnel	Any	None	default
9 · Interop	Any	pymatgen, RDKit	science
10 · Training	Fine-tune MACE on a dataset	torch, mace-torch	mace
11 · AI Agent	Drive TrainCraft in plain language + visualise on a headless VM	Pi agent + Gemma	any