%20copy.png)
Sohrab Haghighat: In the long term, our aspiration is to turn this into the solution that when you want to design hardware, you're going to use. The same way that, for example, today, people use, say, Cloth or Codex for software development.Jordan Metzner: Built This Week, breaking it down. Built This Week, we show you how. A fresh idea, a clever tweak, you locked in true. BuiltSam Nadler: Hey, everyone, and welcome to Built This Week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder here at Rise Labs. And each and every week, I'm joined by my friend, cohost, business partner, Jordan Metzner. How you doing today, Jordan?Jordan Metzner: Hey, Sam. How's it going? Another exciting episode. It's been another crazy week in AI. I think we've got some new releases from both Anthropic this week, as well as OpenAI, as well as Grok.So definitely not a slow week in the AI world, and just lots going on in the AI world in general. But, yeah, super excited for this week's guest, actually.Sam Nadler: Super excited about this week's guest. We've got a great one, and, we'll introduce him here in a second. But before I do, please like and subscribe. We have new episodes out every week. We're about to hit 40,000 subscribers on YouTube, so don't forget.And with that, I'd love to intro our guest this week. Sohrab, the CEO of Hestus, who's developing a really cool product in the Autocomplete CAD space, probably great for all different types of designers, architects, etcetera. But I'm not doing it justice. So, Rob, please introduce yourself. Tell us a little bit about where you came from and a little bit about your product, and then we'll jump into the demo.Sohrab Haghighat: Hi, Sam and Jordan, and thank you for having me. Excited to be here. So I'm Sohrab. I'm an aerospace engineer by training. So, what brings me to the world of AI is years of working at the forefront of software and hardware development and then heartbreaks and pains that I endured myself and I had my team back when I was running my Rocket Launch company enduring when developing hardware, when bringing hardware to life.And all those journeys and mishaps, brought me and my cofounder Kevin to the point that there gotta be a better way, and we are, on a journey to make that better way a reality. So that's in short who we are and what we...Sam Nadler: Perfect. I mean, we're gonna zoom out a little bit into Hestus, but just while we got you, can you show us a little bit of how it works and what it actually does?Sohrab Haghighat: Sure. Exactly. So as you said, what we offer today in form of product is what we call Autocomplete for CAD. So on the left hand side, I'm showing you how a human goes and applies, center lines to a newly designed rectangle. And Hestus knows it the moment it sees the rectangle that, okay, chances are they wanna have some reference lines to reference and then maybe, use that to position the rectangle better.Automatically jumps to prompted to you in a red graphical overlay and the user can say yes that's appropriate to what I'm doing and accept it with a single click or ignore that and move forward. On the right hand side we saw that when the user intended to do that, how they manually have to go that today, it takes far more time and many more manual clicks to do it. And that's just a very, very simple example. It's nothing... It's not a product that I'm showing you how it is going to be different.But along the way, when you went to get to that product, there are many instances of that.Jordan Metzner: So is is Hestus living as a as a plug in inside of AutoCAD?Sohrab Haghighat: So Hestus lives as an extra layer on top of CAD. It could be any CAD that you're using. We are bringing intelligence to your CAD. Exactly a good question. We believe, people have huge corpus of data and designs in the cat that the organization is adopted and are using, and bringing them to a new software is not the right approach.But bringing intelligence to the software that you're using is in fact the right approach. We insist on meeting our users, meeting engineers where they are as opposed to inventing a new workflow for them. Now you want to add for mounting circles to this, for example, backplate that you're designing. So initially it has this thing said, oh, you may want to have a center line for this circle. That's not appropriate.There are always multiple recommendations and you can find the next one and voila, you just cycle through the recommendations and you see that the next one is what you want immediately have it replicated that pattern for you apply all the symmetry requirements and constraints. So it sits locked in whereas again, you want to do it manually, it takes far many clicks to do it manually. You get it done faster. You get it done with few manual clicks. Ultimately, every click that we save for a user is an opportunity of mistakes that we are eliminating from the design, and we get them to finish the design faster.So that's what the product is today.Sam Nadler: Does it also begin to learn behaviors of the user where, you know, maybe that circle they're replicating four times and you mentioned they're... They can, you know, sort through the recommendations and, you know, maybe initially it's the second or third recommendation that is the, acceleration recommendation that they're looking for. But over time, it's like, okay. It always chooses the third one. I'm gonna move that first.Is it is it also learning their kind of behavior with similar projects potentially?Sohrab Haghighat: Yes. On the fly, it has the ability to learn your preferences within that design session. So we are not going to basically extend what we learn and how we adapt the order of recommendations within one session to another session because your choices could be, based on the design needs. So over time, those changes are going to affect the model training that is going to land for everyone. But within each session, depending on recommendations that you choose or you pass, it tries to basically adjust the order to your liking within that design session.Jordan Metzner: Okay. And so when you said... I mean, I I mentioned AutoCAD specifically, but it's CAD agnostic then, you're saying. So, like, any CAD software Hestus can sit on topSohrab Haghighat: of it? Correct. That is the... Basically, the plan. Today, are available on Autodesk Fusion, which is, the leading CAD product offered by Autodesk.We are, in a very basic... In matter of one or two weeks, we are planning to release on SolidWorks as well, which is another leading CAD product. And then over time, we are going to support other CAD platforms such as Onshape, such as NX or CATIA.Jordan Metzner: Okay. Awesome. And just in, you know, your experience since building this, have you found specific industries in which, you know, your software's been as specifically applicable? Has it been more in industrial? Is it in architectural?Is it in, I don't know, maybe in parts design? Kind of where where have your customers found kind of the best implementation and use cases?Sohrab Haghighat: We are primarily focused on hardware design and hardware design applications, less so architecture. However, I should say I have seen some users using Hestus to help them that, basically design of an apartment or house and all those lines that it takes to basically manually add them, manually dimension and apply constraints, it helps them go through it faster. But our target audience is mechanical engineers, not necessarily mechanical engineers. We call everyone mechanical engineers, but an aerospace engineer, a robotics engineer, anyone who is designing hardware. And one of the beauties of how we are approaching, this problem slash solution is that we started from a top...A bottom up approach. Basically, we thought that, okay. We need to build this AI in a way that we went to school and became engineers. And where does every design start from? Starts from a sketch.You need to have a two d sketch to then later add depth to it. And because we started from this very bottom up, methodology, everyone can benefit from it. And we have seen our users come from a wide range of industries, from industrial valve design to designing drones, to designing diesel engines, all the way to people who have used it to design dining chairs. So... And that is what makes me very excited about the application of Hestus.That's super cool.Sam Nadler: Working out is easy to start, hard to stick with. HipTrain fixes that. Real coaches, real accountability, a plan that actually fits your life. No guessing, no skipping, just consistency. Hip train.Jordan Metzner: The output is this compounding time saving of of effort that happens over a design period where starting at x to y used to take, you know, z hours and now it takes one quarter z hours because you have Hestus here helping you out through the process.Sohrab Haghighat: That is the measurable, the tangible benefit of using testes today that we are going to save you manual clicks and in doing so, we are going to help you finish the same job faster with you as the designer in the loop. You have the full control over how the design is going to look like. So it's your design. It's your engineering. We just help you today to finish it fast.In the long term, our aspiration is to turn this into the solution that when you want to design hardware you're going to use. The same way that, for example, today people use, say, Cloud or Codex for software development, and you just don't use it anymore to write a function for you quicker, you brainstorm with it, you try to come up with an architecture, you come up with an execution plan, and then go through every element of it, that is the ultimate goal that we have. A system like that that would allow you to go from concept to a finished product that is ready for manufacturing, and that is the key thing that we want people to understand. There's a significant gap between here's a flashy three d model versus a three d model that is ready for manufacturing, and that is what I call it. We want to help engineers architect that solution, execute steps, and make sure it is going to be ready for the manufacturing methodology that they have in mind.But, we believe in a sustainable approach, so that's why we are going bottoms up. Okay. Let's first sketch then do three d, then do three d assemblies, then bring in manufacturing on it. And once we have that, let's glue it together in an AI that grasps the whole thing.Jordan Metzner: Okay. That's awesome. How how can people learn more, about Hestus and and, and learn about you?Sohrab Haghighat: So number one thing, go to our website, download it. It's available for free. The main reason we are doing this, the reason Kevin and I started Hestus is because we were frustrated as engineers ourselves, and our motto is we want to make engineers' lives better. So go download it. Go start using it.And as I mentioned, available on Autodesk Fusion very soon is going to become available for SolidWorks as well. And as you get to use it, let us know what other features would, help you and help your work, flow better. We do have an icon that looks like literally a bug in the software that you can use it to report, issues that, okay, gave me a bad recommendation, or I was expecting that recommendation. That would have been more appropriate,Sam Nadler: but it was, not one, the top recommendation. So... And we'll get on it. We'll just make sure that you work on it and release it and let you know and... Yeah.Amazing. Amazing. Well, with that, it was a pleasure to hear more about Hestus. With that, I want to introduce a little bit of news for this week. You know, while it seemed like maybe the Frontier Labs were were talking about slowing down, they both released new models, I think, yesterday and today.Anthropic releases 5.5, Lower prices that... And compatible performance to Fable. I believe OpenAI launched Soul and Luna, g p t six, lower cost as well. Curious if either of you... Jordan, let's start with you.Have you had the chance to test any of these models out? Immediate reactions and, yeah, high level thoughts.Jordan Metzner: Yeah. I gotta play with a little bit. I tested Opus, the new Opus model, which is performing quite well and significantly cheaper than using Fable, so that feels pretty good. I also use Codex a lot. So I've been using some of GPT six.I haven't tried the new GPT six. The other models, the lower level models only sold so far, but it's been pretty cost effective. So I haven't really found, like, the need yet. But I think to your point, Sam, you know, it appears that, you know, these labs are not slowing down. Like we mentioned at the beginning of the show, you know, Brock released a model this week.It doesn't seem like the Twitter audience or Rated audience has picked it up as, like, you know, working as successfully yet. But, you know, I don't think you can count on Elon and, you know, the cursor team over there. So I think we'll probably see more there. Google continues to release more, like, real time models with TTS and, like, 3.8 Gemini. So I think all of them are getting more and more competitive.And then, obviously, you know, we have what's happening in open source. Right? So, you know, it's a lot of exciting stuff, but I think the overall point is is that intelligence is getting cheaper. Right? 5.5 when Opus came out was incredibly expensive.Right? And then Fable's incredibly expensive, you know. And and now we've, you know, dropped it down, you know, to, you know, significantly lower cost per token prices. So that feels good, and the models are really stable. I don't know.What about you guys?Sohrab Haghighat: Yeah. So personally, I use anthropic models for, programming, myself. And, yes, it is exciting that the cost of intelligence is going down. That's, always worth celebrating for. However, I do agree with you, Jordan, that there...As much as there are high level talks about we should slow down, there doesn't seem to be any of that talk really materialized. For our own development, so, obviously, the more advanced models that we can use to faster build our own product better. The the advance in l and m's necessarily is not going to materialize in form of advancement in our model because we believe, to solve the problem for physical design, we need to have a custom architected model built from ground up meant for that specific application. So there's not a direct translation of LLM improves. That means that Hestus is going to improve.LLM improves means that we can code faster to build our model fast, but, which is always something that we're up for it. If it's cheaper for us to code, amazing. That's great.Sam Nadler: So, Rob, it was a pleasure. Thank you for joining. This may be the episode that kicks us over 40,000 subscribers, so that will be a a big milestone for us. Everyone, go check out Hestus. It looks like a great product, that's accelerating people's ability to design hardware faster.So, Jordan, any final thoughts?Jordan Metzner: Great episode. Great product. So thanks for the demo. That was really cool. And, yeah, great episode.And thanks, Sam. Thanks, Rob. See you guys soon.Sohrab Haghighat: Thank you for having me here.Jordan Metzner: Awesome. Thanks, Rob. Bye, guys.
