Home › Forums › Startup Investors › Offering Questions and Discussion › Looking at a company and would like some expert input › Reply To: Looking at a company and would like some expert input
OK – I took a look at the correct Guru (by MSBAI) this time. Let’s try this again.
From what I gather (took me a little while reading their page), they are essentially trying to:
- Allow generalist engineers to run more specialist simulations
- Allow these engineers to use High Performance Computing (HPC) for their simulations (essentially running it on supercomputers and spreading the simulation among a bunch of processing cores and/or GPUs)
Is it a good idea? Sure – enabling more generalist engineers to carry out the task of specialists would help lower overall costs for running these types of simulations and models…assuming they are done right, and that’s a big assumption. There is a reason today that specialists are typically required for these types of things.
Now, is it a practical idea? That needs some additional questions answered…
First, I’m speaking purely from an aerodynamic and thermal simulation perspective for electronics and airplanes/helicopters – I can’t provide any input really on the Energy Efficiency side. I used to run Computational Fluid Dynamics (CFD) analyses using HPC on our own supercomputer grids of 64-128 processors.
I personally think this could be useful for small businesses (and even generalists in larger corporations), but primarily for running first-pass analyses during preliminary design and while conducting trade studies. I think if you’re going to manufacture something complex like an aerodynamic fairing, you need the expertise and control of a specialist. Or you at least need the assurance that Guru is using the proper assumptions. It’s great that it abstracts some of that complexity, but a common saying is “garbage in, garbage out” – you can get any type of beautiful results, but if you didn’t set it up properly, the results will be meaningless (although they’ll still look pretty).
If I had to guess, because the CEO/Founder seems to have an in-depth aero background, I’m guessing their “AI” today is mainly taking their own knowledge and expertise and productizing that expertise via the Guru interface. One of my main concerns is that they tout “AI” as being a major part of this product, yet right now it seems neither of the founders (nor advisors) have in-depth AI experience, and I question whether you could pull it off by outsourcing that part of the product development to someone else.
Personally, I also don’t get the voice command feature – it’s either a very minor feature that they are playing up in the marketing, or I question whether they are spending time on where the real value is to be added. Maybe I’m just set in my ways and will be proven wrong, but I find it hard to imagine a room full of engineers giving simulation commands via voice.
Questions I have about the campaign page:
- All their Letters of Support are from early 2018 – any updates since then?
- Hard to tell if the team is full-time – both have other jobs/professions listed on their LinkedIn page
- Who is the Artificial Intelligence expert? AI experts cost huge $$ in today’s market and I don’t see either of them with solid AI backgrounds, so I question their ability to pull of the AI part of the business
- Maybe I don’t get the product, but I personally can’t understand why they are focusing on a voice interface – seems like more of a sales gimmick to me. Any engineer looking to run this would have some type of decent computer set up.
- BIG questions about the pricing and unit economics, especially if they plan to target Small and Medium-Sized Businesses (SMBs). They mention $500 per simulation or $20,000 per year, but this depends on a lot of assumptions – are the solvers they are using open source or commercial? Does it depend on compute resources requested? How does protecting sensitive data / restricted data work with the cloud computing, etc. See questions below
I had some more technical questions about how the product would work, which I posted to their campaign page. Answers to these questions would help address whether it will really be as cost-effective or not as they are saying for SMBs.
Here are the questions I posted to them on their campaign page:
- Does the company still have to own or lease the software license(s) for the software simulation package, or is that included in Guru’s cost? (e.g. if I wanted to run a FLUENT CFD simulation on 32 cores, do I need FLUENT or HPC licenses)? Or do you only use open-source solvers?
- Do users pay based on resources used or time, or are there caps on what is included in the standard Guru subscription? e.g. if I want to run on 128 cores instead of 64 to get quicker results, am I charged a different amount?
- While it would be ideal to say “Guru, go run this geometry at these conditions” and have it spit back results, there will always be a need for adjustments in settings, checking the grid quality, etc. Does Guru allow for these adjustments and checks along the way?
- What about sensitive data and simulations, especially for DoD, DoE, and others who might have ITAR and other restrictions? Does Guru still allow you to run on local clusters, and does the supercomputer implementation today allow for handling sensitive (but unclassified) data?
Hope this helps! If I somehow managed to review the wrong one again, feel free to call me out 🙂