MatSX (2025 KRICT ChemDX Hackathon)

MatSX is an AI-powered platform for similarity search and anomaly detection in materials. This repository is a prototype project developed during the 2025 KRICT ChemDX Hackathon.

App demo

AI Agent Demo

AI Agent demo video here.

🏆 1st Place Winners! 🏆

We are proud to announce that our team won 1st place at the 2025 KRICT ChemDX Hackathon!

The winning team celebrating our victory at KRICT 2025

Team Members

• Seongho
• Seonhye
• Youngtak
• Hyunsoo
• Ry

---

ChemDX Explorer with AI-Assisted Materials Analysis

Project Overview

This repository contains a comprehensive materials science platform that combines interactive visualisation with AI-powered analysis. The system integrates Materials Project and MatDX databases to provide advanced materials discovery capabilities through multiple interfaces:

• Web Application: Interactive 3D visualisation with Dash
• AI Agent: Intelligent materials analysis with OpenAI integration
• Command Line Interface: KRICTHACK for rapid materials search

Core Features

Search API (src/search_api.py)

• Computes embeddings (Magpie for composition, MACE for structure) from material inputs.
• Applies dimensionality reduction methods (PCA, UMAP, Adversarial Autoencoder).
• Identifies similar materials from the Materials Project database by calculating Euclidean distances in the reduced embedding space.

Web Application (app.py)

• Interactive 3D scatter plots for materials space exploration
• Crystal structure viewer with nearest neighbour display
• Support for multiple embedding methods (Magpie composition-based, MACE structure-based)
• Dimensionality reduction techniques (PCA, UMAP, AAE)

AI-Powered Analysis (KRICTHACK)

• Intelligent similarity search for materials
• Natural language queries for materials discovery
• Interactive chat mode with conversation memory
• Integration with Materials Project API for real-time data

Search and Embedding System

• Pre-computed embeddings from H5 files for rapid search
• Support for both composition and structure-based similarity
• Advanced embedding techniques including adversarial autoencoders

Quick Start

Environment Setup

Create a .env file and add the following content:

MP_API_KEY='your-materials-project-key'  # For materials project
OPENAI_MDG_API_KEY='your-openai-key'  # For AI analysis

Running Applications

Web Interface

# Main Dash application (port 8050)
python app.py

# With custom settings
python app.py --debug --port 8080

AI Agent (KRICTHACK)

# Search for similar materials
kricthack search "Fe2O3"
kricthack search "battery cathode materials"

# Interactive chat mode
kricthack chat

# System diagnostics
kricthack setup

Acknowledgements

Great thanks to KRICT for hosting the hackathon and providing the research opportunity.