Kudos to the authors: Madeleine A Gaidimas, Abhijoy Mandal, @chenpanxyz.bsky.social , Shi Xuan Leong, Gyu-Hee Kim, Akshay Talekar, Kent O Kirlikovali, Kourosh Darvish, Omar K Farha, @variniabernales.bsky.social, @aspuru.bsky.social
[6/6]

Why it matters:

By integrating computer vision into automated, high-throughput workflows, experimentalists can extract rich information from visual data already being generated—accelerating materials discovery and reducing reliance on manual, low-throughput characterization.
[5/6]

🧪 A multi-institutional case study tracking MOF crystallization in a high-throughput synthesis platform
💬 A user study demonstrating measurable gains in speed, insight, and reliability from AI-driven characterization
[4/6]

What you’ll take away:

📷 How to build CV-enabled characterization pipelines (hardware + software)\
🤖 How to train, validate, and deploy CV models in real experimental workflows
✏️ Practical guidance on image acquisition, annotation, and data quality
[3/6]

Modern synthesis platforms generate vast amounts of visual data, yet much of it is underutilized.

Written by and for experimentalists, this tutorial provides a practical, lab-ready roadmap for integrating computer vision (CV) into materials workflows - no prior CV or ML expertise required!
[2/6]

- Hossein Darvish, @allanzhao.bsky.social, @gkwt.bsky.social, Han Hao, Miroslav Bogdanovic, Gabriella Pizzuto, @aicooper.bsky.social, @aspuru.bsky.social, Florian Shkurti, @animesh-garg.bsky.social.

[6/6]

Kudos to all the authors: Kourosh Darvish, Arjun Sohal, Abhijoy Mandal, Hatem Fakhruldeen, Nikola Radulov, Zhengxue Zhou, Satheeshkumar Veeramani, Joshua Choi, Skyler Han, Brayden Zhang, Jeeyeoun Chae, Alex Wright, Yijie Wang -

[5/6]

Why it matters

💻 By enabling in-silico testing of automated workflows, MATTERIX reduces reliance on expensive and time-consuming physical testing, accelerating materials discovery and lab automation at scale.

[4/6]

Key capabilities:

🤖 Robot skills and hierarchical planning
🧂 Powder & liquid simulation
🧠 Modular semantics engine for simulating physical and chemical processes such as heat transfer, pH changes and many more
🧱 Open-source asset libraries
🎯 Demonstrated sim-to-real transfer

[3/6]

🚀 What’s MATTERIX?

MATTERIX is a GPU-accelerated simulation framework for building high-fidelity digital twins of chemistry laboratories. It bridges the gap between workflow design and robotic execution.

No programming background? MATTERIX can set up complex workflows with just a config.

[2/6]

Also, check out the latest addition to the El Agente family, El Agente Cuántico: arxiv.org/abs/2512.18847
[5/5]

Below, we share links to the agentic systems featured in the article, alongside El Agente ( elagente.ca ):

🕶️ El Agente: www.cell.com/matter/fullt...
⚛️ Aitomia: arxiv.org/abs/2505.08195
🧪 Chemgraph: arxiv.org/abs/2506.06363
🧠 DREAMS: arxiv.org/abs/2507.14267
[4/5]

By offering LLM-driven natural-language interfaces, these agents are greatly reducing the barrier to entry for chemists who are not yet trained in the syntax of advanced computational chemistry tools.
[3/5]

In it, our directors @aspuru.bsky.social and @variniabernales.bsky.social share their perspectives on the subject, along with insights from the Matter Lab and the El Agente team.
[2/5]

Across benchmarks covering both smooth and rugged chemical landscapes, rank-based surrogate models often match or outperform standard regression approaches, especially early in the search and in challenging settings with activity cliffs.
[3/3]

In many discovery problems, what matters most is which molecules are best, not the exact value of their predicted properties. This post argues for a simple but powerful shift:

💡 Train Bayesian optimization surrogates to rank candidates, rather than predict precise property values.
[2/3]