Atomcraft v0.1.0 — AI-Powered Materials Discovery Platform

Atomcraft v0.1.0

AI operating system for accelerated materials discovery. From prompt to synthesis-ready material.

What's Included

• 104,842 Materials Project entries pre-loaded
• 11 trained ML models (RandomForest) for property prediction
• 112-class space group classifier (95.9% accuracy)
• Materials generator (de-novo + template-based substitution from 68-element pool)
• Synthesis engine (5 methods: solid-state, sol-gel, hydrothermal, CVD, mechanochemical)
• DFT input generator (VASP, Quantum ESPRESSO, CP2K, LAMMPS)
• Active learning loop (pseudo + DFT mode)
• Knowledge graph (Neo4j integration)
• Conversational AI interface (GPT-4o via LangChain)
• Full React frontend (Dashboard, Materials DB, Generator, Predictor, Experiments, Chat)
• 3D crystal viewer (Three.js)
• Docker Compose deployment (5 services)
• RESTful API with 25+ endpoints

Quick Start

# Docker (recommended)
docker-compose up --build

# Local
python3.13 -m venv .venv
source .venv/bin/activate
pip install -r backend/requirements.txt
uvicorn backend.app.main:app --reload --port 8000

Architecture

Backend: FastAPI + Celery + SQLAlchemy | Frontend: React 18 + Vite 5 + Tailwind + Three.js | ML: scikit-learn RandomForest | Infra: PostgreSQL + Redis + Neo4j

Full docs: https://github.com/fridayowl/atomcraft/blob/main/README.md