---
title: MEIDNet Prism
emoji: 🔮
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 7860
pinned: true
license: mit
short_description: Learn, build and benchmark multimodal AI for materials
thumbnail: https://babu09-meidnet.hf.space/docs/assets/meidnet_prism_logo.png
tags: [materials, inverse-design, generative, crystal, perovskite, multimodal, contrastive-learning, chemistry]
models: [Babu09/MEIDNet]
---

<p align="center"><img src="https://babu09-meidnet.hf.space/docs/assets/meidnet_prism_logo.png" width="640" alt="MEIDNet Prism"></p>

# MEIDNet Prism

**Learn, build and benchmark multimodal AI for materials discovery.** MEIDNet, the reference
implementation, learns one shared latent space for crystal structures and their properties, enumerates a
prototype-family design space checked by chemistry rules, and searches the latent space for candidates
with the properties you want.

* **[Home](https://babu09-meidnet.hf.space/)** — overview and a guided tour, from concept to demonstration.
* **[Learn](https://babu09-meidnet.hf.space/docs/learn/index.html)** — what a modality is, the five
  challenges of multimodal learning, contrastive learning with a playground, the
  [Architecture Atlas](https://babu09-meidnet.hf.space/docs/learn/architectures.html) with an advisor,
  how MEIDNet works, and [the ecosystem](https://babu09-meidnet.hf.space/docs/ecosystem.html) it fits in.
* **[Benchmark](https://babu09-meidnet.hf.space/docs/benchmarks/index.html)** — leaderboards under fixed
  protocols (inverse design, property prediction, representation on Perov-5, with baselines), metric
  families named as in LeMat-GenBench plus a conditional extension, `meidnet score` for any model's
  generated structures ([compatibility](https://babu09-meidnet.hf.space/docs/benchmarks/compatibility.html)),
  datasets and databases, and [contributed results](https://babu09-meidnet.hf.space/docs/community/contribute.html).
* **[Develop](https://babu09-meidnet.hf.space/docs/develop.html)** — the `meidnet` package (`pip install meidnet`),
  the [Studio](https://babu09-meidnet.hf.space/studio/) with the workflow *Data → Model → Family → Rules →
  Targets → Search → Candidates*, your own data, recipes, the configuration file, the Python API.
* **[MEIDNet Matter ↗](https://babu09-meidnet-matter.hf.space/)** — the companion application for
  researchers: bring a dataset, set a design goal, read whether the data and model support it, search for
  candidates with evidence next to each one, export them with a validation ladder.

**Model:** [Babu09/MEIDNet](https://huggingface.co/Babu09/MEIDNet) (the pretrained Perov-5 checkpoints) ·
**Code:** [github.com/ABnano/MEIDNet](https://github.com/ABnano/MEIDNet) (MIT) ·
**Paper:** A. Babu, R. A. Gouvêa, P. Vandergheynst, G.-M. Rignanese, *npj Computational Materials* (2026),
[doi:10.1038/s41524-026-02153-3](https://doi.org/10.1038/s41524-026-02153-3).

**Further reading**

1. Y. Bengio, A. Courville, P. Vincent, *Representation learning: a review and new perspectives*, IEEE TPAMI 35,
   1798–1828 (2013), [doi:10.1109/TPAMI.2013.50](https://doi.org/10.1109/TPAMI.2013.50).
2. Y. LeCun, Y. Bengio, G. Hinton, *Deep learning*, Nature 521, 436–444 (2015),
   [doi:10.1038/nature14539](https://doi.org/10.1038/nature14539).
3. B. Sanchez-Lengeling, A. Aspuru-Guzik, *Inverse molecular design using machine learning: generative models for
   matter engineering*, Science 361, 360–365 (2018), [doi:10.1126/science.aat2663](https://doi.org/10.1126/science.aat2663).
4. A. Babu, R. Almeida Gouvêa, G.-M. Rignanese, *Toward automated discovery with generative models multimodal
   learning and closed loop workflows in inverse materials design*, Cell Reports Physical Science 7, 103561 (2026),
   [doi:10.1016/j.xcrp.2026.103561](https://doi.org/10.1016/j.xcrp.2026.103561).
5. A. Babu, N. M. A. Krishnan, *Multimodal and cross-modal learning techniques*, APL Machine Learning 4, 030901
   (2026), [doi:10.1063/5.0346744](https://doi.org/10.1063/5.0346744).

This public Space is shared: uploads are capped (8 MB, 1,500 rows, 30 epochs), one training at a time,
sessions are private per browser tab and removed after an hour. Predicted properties are the model's
estimates — confirm candidates by DFT or experiment.