The motivation behind this project arises from a very general question that we constantly face in our daily work as engineers and scientists. Additive manufacturing (AM) offers unprecedented design freedom, but it also poses a fundamental challenge: how do we decide which design is optimal for a specific process or task? In our case, we aim to discover, design, and manufacture the next generation of structured devices for chemistry and chemical engineering applications, both in academia and industry.
When we began exploring the design of structured reactors for continuous-flow chemistry, periodic structures—such as grids (https://doi.org/10.1039/D1GC04593H - Green Chem, 2022), helicoidal designs (https://doi.org/10.1016/j.addma.2024.104304 - Additive Manufacturing, 2024), or gyroid-based geometries—naturally became our starting points. In particular, gyroids offer high surface area and tortuosity, which, combined with their mathematically elegant complexity, made them appear as an ideal solution for performance optimization.
But soon we began to wonder:
Is the gyroid truly the best geometry for every reaction?
Or were we relying on a default topology without knowing whether it was optimal for each mechanism, phase regime, or kinetic pathway?
This question sparked a broader idea:
if we could generate many geometries, fabricate them rapidly, evaluate them automatically, and learn from the resulting data…
could we design a reactor “tailored” to each specific reaction?
That is how Reac-Discovery was born, a platform that integrates mathematical geometry generation, 3D printing, and machine-learning-based optimization.
WHAT IS REAC-DISCOVERY IN SIMPLE TERMS?
Reac-Discovery is built on three modules:
- Reac-Gen
Reac-Gen is a software tool developed by our group to generate POC (Periodic Open-Cell) structures. It is built around a library of 20 distinct families of geometries, each defined by implicit equations and fully parametrizable, enabling the exploration of millions of possible structural variations. In addition, the software computes key geometric descriptors, including tortuosity, surface area, free volume, and porosity.
2. Reac-Fab
Reac-Fab transforms digital geometries into physical reactors using MSLA (Masked Stereolithography) and acrylate-based photocurable resins developed by our group.
A neural network predicts whether each structure will be printable before fabrication. This can be expanded to other AM techniques.
3. Reac-Eval
Reac-Eval is a self-driving lab platform, which autonomously controls process parameters such as temperature, pressure, liquid and gas flow rates, and concentrations. It analyzes and records in-line data, evaluates multiple structures, and trains models that correlate geometry, process conditions, and performance.
In principle, this platform can simultaneously optimize both process conditions and reactor geometry in a way that has not been possible before. We hypothesized that Reac-Discovery would deliver a tailored response to the different chemistries processed through the platform, and this was validated with two distinct reactions: acetophenone hydrogenation and CO₂ cycloaddition. Following optimization, Reac-Discovery identified a customized structure for each case. Moreover, this approach allowed us to achieve a record space–time yield (STY) of 803 g L⁻¹ h⁻¹, the highest reported under triphasic conditions with immobilized catalysts.
WHY AN ANIMATION?
Beyond the experimental results, we wanted to convey the full essence of the platform:
a continuous flow in which experimental data feed the models, the models generate geometries, 3D printing materializes them, and molecules ultimately travel through reactors designed specifically for each reaction.
No conventional technical schematic could capture this interplay between information, design, and fabrication. We therefore chose to develop a conceptual digital illustration brought to life through animation, capable of communicating the entire story of the project at a glance.
THE MEANING BEHIND THE VISUAL NARRATIVE
The animation was created by Carmita Salgado (https://cvsstudios.art/), an emerging scientific illustrator, and it became a visual synthesis of the whole Reac-Discovery process.
A. The structured reactor at the center
The central element is a POC reactor generated with Reac-Gen. Its periodic cavities evoke the mathematical geometries we use to optimize transport phenomena. It represents our starting point: turning equations into functional geometries.
B. The rain of numbers and symbols
This “rain” symbolizes the experimental data produced by Reac-Eval, translated into features and targets that feed the neural network responsible for learning correlations and proposing new designs.
Key detail:
the numbers correspond to 8-bit binary codes spelling out “Reac-Discovery.”
A direct metaphor for how information transforms into design. For instance, 01010010 (“R”) and 01100101 (“e”) appear among the falling digits, subtly encoding the word “Reac-Discovery”.
C. The neural networks beneath the reactor
The interwoven lines represent the ANNs trained with real reaction data.
These networks:
- learn relationships between geometry, conditions, and performance
- identify patterns invisible to human intuition
- and generate reactor architectures tailored to each target reaction
They are the cognitive engine of the platform.
D. Molecules traveling through the reactor
CO₂ and epichlorohydrin enter on one side, and cyclic carbonate exits on the other.
This represents the molecular journey of one of our case studies and reinforces that Reac-Discovery designs geometries for real chemical systems, not in abstraction.
E. The photocurable resin and the UV light below
At the bottom is a pool of acrylate-based photocurable resin (developed by our group), from which the reactor geometry is physically created. The bright light represents the UV projector of an MSLA printer curing the structure layer by layer.
The reactor appears upside down, exactly as it would be printed in MSLA.
This element highlights not just the design, but the physical act of fabrication.
F. The complete visual narrative
The vertical composition narrates the entire workflow of the project: digitized experimental data feed the neural networks that enable the design and optimization of both process parameters and reactor topology. The reactor then materialized as a functional solid, specifically tailored to each target reaction, via MSLA.
In the illustration, you can see the up-flow journey of the molecules through the reactor and their transformation as they pass through it, showing how Reac-Discovery can model the behavior of any reaction and drive the global optimization of the system.
The image allows the viewer to grasp the core of the project without resorting to equations or technical plots. It shows how data, mathematics, 3D printing, and catalysis are integrated into a coherent flow. It acts as a bridge between the digital and material worlds, which is precisely the essence of Reac-Discovery.
WHAT COMES NEXT?
Reac-Discovery is only just beginning.
Our vision is to expand the platform beyond these initial case studies, incorporating new materials, exploring more complex geometries, and opening the door to a future where reactors are designed not by intuition but by evidence, data, and machine learning.
We also envision this methodology empowering other laboratories, even those without expensive infrastructure, to generate high-performance solutions through accessible digital design and additive manufacturing.
If the future of chemistry is autonomous, reproducible, and data-driven, this work represents a small but meaningful step in that direction.
ACKNOWLEDGEMENTS
I would like to acknowledge everyone who made this project possible:
my team supervised by Professor Víctor Sans, my co-authors for their collective effort, and Carmita Salgado (https://cvsstudios.art/), whose illustration does not simply accompany the paper but amplifies it, making it more human, accessible, and visually memorable.
It has been a privilege to work with them.
By Cristopher Tinajero and Victor Sans