TecunTecs Logo

TecunTecs (Proof of Concept)

YC Demo Fall 2024

Join Beta Waitlist

No-code App for Cloud-native Physics-informed AI

Aneurysm Geometry

Simulation Configuration

3D Scatter Plot of Pressure Data

Run NVIDIA Modulus on AWS EC2 GPUs— Physics Informed Neural Networks — STL Geometry: Aneurysm Blood Flow


The purpose of this demo is to demonstrate how a no-code app for physics-informed AI could simplify complicated workflows for domain experts.


To illustrate, here is an excerpt of the description of the YouTube tutorial below:


In this tutorial, I'll guide you through setting up an AWS EC2 instance, installing NVIDIA Modulus, and running a full-scale blood flow simulation. We’ll start by preparing your AWS EC2 environment and walking through the installation of NVIDIA Modulus. You'll learn how to handle the complex geometry of an aneurysm from STL files and configure the simulation to analyze blood flow and pressure in this critical area.


One of the challenges of this project is dealing with the computational demands. I’ll show you how to set up an NVIDIA GPU-Optimized AMI on AWS to handle this, explaining why running this on a Google Colab might not be feasible due to limitations, and how the EC2 instance, though costly, offers the necessary resources for such an intense simulation.


After running the simulation, we’ll cover how to extract and download the output files for post-processing. You’ll learn how to load model weights, visualize the pressure and velocity fields, and use TensorBoard for an in-depth analysis of the simulation’s progress.


If you pay attention, you'll notice that it takes a lot of work to do the set up, run the simulation, post-process, and analyze the results. This is where a no-code app can help. By abstracting away the complexity of the setup and running the simulation, domain experts can focus on the results and the insights they provide.

NVIDIA Modulus Crash Course — Physics Informed Neural Networks (PINNs) — 2D Lid Driven Cavity Flow


An ideal customer profile (ICP) is an engineering simulation user (of tools like Ansys) who wishes to leverage Physics-informed Neural Networks for their advantages such as transfer learning that enable to repeat similar simulations faster and at a potentially reduced cost. I've also been told by this kind of users (people have reached out via LinkedIn after watching my videos) that they wish dealing with simulation software (e.g., Ansys), or even the implementation of PINNs themselves, was more user-friendly.

Me Explaining the Demo!


Why do I need the money? Time & Freedom


Time is more valuable than money, and freedom is more valuable than time. (Would you rather live a slave or die a free man?)


The hardest thing I'll overcome in my life is permanently having full rights to stay and work in the United States. I was rejected for an EB-1 visa/green card ("Alien of Extraordinary Ability") in 2020 (I had just turned 22). If I get funding for this startup, I'll apply again so that I can be free to move to the US. With premium processing, I should be able to get there fast. I won't give up.


Who am I and what I believe in?


I am José Andrés Alvarez Cabrera and I believe that freedom of movement is an unalienable right.

Join me at jose at tecuntecs dot com


"Good video, nicely explained! Hope you can help as much people from the scientific community as you can!" — @BenjaminAlfredoTourn (youtu.be/XlCGi_DCJIU)


"we wait for the docker installation keep up the good work" — @user-of6vf9uv3z (youtu.be/XlCGi_DCJIU)


"Nice video, thanks for sharing" — @godswillotto9110 (youtu.be/XlCGi_DCJIU)


"I came for copper and found it was gold." (Translated by Google) — @wilsonandresmorenoreina5896 (youtu.be/te39qrgQ-Ao)


"Thank you for explaining it clearly. Can you please explain us how to do post processing for 3d fin problem" — @bhargavguntreddi6032 (youtu.be/te39qrgQ-Ao)


"Hi, thanks for sharing. this is very helpful. I am very new to pinns and i keep getting an error 'RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)' not sure how to fix this. Have you ever ran into this?... made it work!!!!! thanks for this videos. i cant wait to watch more and learn... one more question. do you think it is possible to use this model to check a site plan? with multiple 2d "squares" or geometries?" — @AndresRoncal (youtu.be/te39qrgQ-Ao)


"Hi, the link you shared the colab notebook is not working, can you update it(make it public). It would really help." — @egekara6951 (youtu.be/te39qrgQ-Ao)


"Hi mate, thank you for this nice introduction to solving NS using Nvidia Modulus. I made my master's degree thesis using Modulus, but I found it hard to solve even simple 2D cases like an airfoil in a uniform flow. Have you ever tried something outside the examples given by Modulus developers, it would be fascinating for me to have a comparison with someone else. Thank you again for your work :)" — @paolovisentini9661 (youtu.be/te39qrgQ-Ao)


"I am actually currenlty doing my masters and modulus would be a core part in solving 2D and 3D cases. Would like to hear your experience." — @brenalddzonzi7334 (youtu.be/te39qrgQ-Ao)