
Automating design for the physical world | Drones and 3D printing | Building @kyrall_
München, Bayern
Joined July 2023
- Tweets465
- Following919
- Followers788
- Likes536
Today, we're officially launching Kyrall
Until now, AI models couldn't capture design intent in engineering.
We're changing that.
Upload specification documents, sizing tool results, requirements, and past designs, and Kyrall builds a Knowledge Graph, so it not only understands what you work on, but how you work.
You simply describe what you want to design, and Kyrall generates a fully parametric assembly, grounded in your engineering knowledge.
Not a mesh. Not a rendering. Real, editable, parametric 3D models.
you are born and raised in a closed room your whole life, you are an expert mech engineer who knows everything about designing, simulating and testing hardware. However, you have never built any machine in your life, all of a sudden the door of the room opens and someone brings in a built machine that you designed.
The question: did you learn something new in this moment about mech engineering?
8
Even if you are early on something, you need to keep reinventing what being early means. Once the tourists catch up, your “early” ideas become mainstream and you fade into the new crowd.
The trick about having insights in a niche domain is that you continuously have new insights in that domain. It is the delta between the new ideas and the crowd that give your ideas value
1
1
139
So now we are getting at least 3 Emails a week from people selling us CAD data. Two patterns emerge:
1. Completely outdated data
2. Simple machines and parts
While this is better than nothing for training, it is still far away from the type of data needed to have AI design and validate assemblies for proper production (Mainly because production and testing data is missing) @emm0sh talked about this several times.
Also, I think current AI models are already capable of designing most of those simple parts being sold for +$100k.
I wonder if all those brokers are collecting and buying data based on true demand or in anticipation of it.
Feels like the latter as the data they are offering seems random and unstructured. If someone was asking for data, you would assume they will ask for specific formats and structure. Also, if someone was asking for data they wouldn’t be selling it to us.
So, the problem is still there, “no proper mech datasets exist.” I still think the solution is simpler than most people think.
3
4
473
So much value to be captured in AI for engineering, probably more than most other sectors considering the lack of innovation that has been a trend for over 20 years
introducing Gemini 4 Neon: frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows
priced at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window
try it today via the Gemini API and in AI Studio ai.studio/build
8
374
Had a great conversation last night with a VP at a major company who’s spent 15 years building consumer products.
His take on why nothing has replaced the smartphone: thumb movement is one of the lowest-energy, very natural, and highest-dexterity movements we can make.
We spent a long time on glasses that project a screen you control by voice. Where we landed: you still need a touch surface (basically a phone screen) for all the small adjustments voice can’t handle.
The logic is simple. Visual input carries far more information than voice or any other modality. Try teaching someone to cook by only telling them how (possible), versus showing them (much easier). Same with operating a machine or learning a movement. At the same time, exact movements and commands are still needed and can be replaced with voice only for some use cases.
Maybe Steve Jobs “ruined” tech by building something close to the perfect consumer product.
3
243
One interesting finding from talking to people who lived through the shift to digital drafting and then 3D modelling in the 1980s and 1990s is the how little retraining was done and how many drafters and engineers objected to CAD in contrast to teaching junior engineers the new tools. Maybe something to learn there.
At one point the job of the drafter vanished and technical drawings came directly from the 3D model as they are standard now.
curious to learn more if anyone has insights or think what I said is not true
CAD software won’t vanish soon, if anything I predict it will get more accessible and cheaper, here is why:
Don’t be surprised if the CAD vendors start open sourcing their long closed data formats and software.
The reason?
With the increased ease of developing enterprise level software, users will expect from the CAD vendors that they justify the high cost of the licenses they use. Users won’t accept the closed source formats and software anymore and will start expecting more openness as the switching cost gets lower.
Enterprise integration will remain the main bulk of the revenue of the major CAD vendors and I think open sourcing many of their products won’t cannibalise their profit as they increase their margins through new automation tools that will be usage or outcome based rather than the classical license model.
It is interesting to see the discussion about AI in Engineering move from designing one signal part that fulfils requirements to designing a whole assembly.
I predict the discussion after that will be shifting from geometric design (design a drone frame with XYZ airfoil, span, chord length, etc.) to functional design (design a drone frame that has XYZ lift to drag ratio) i.e. describing the function rather than the geometry.
2
12
477
The solution to this; and generally the (lack of) data problem; seems to be very obvious to me.
I wonder why no one (at least openly) is working on this…
this is exactly why assembly CAD is not verifiable and why LLM CAD will plateau after single parts
no amount of FEA, CFD or XYZ will get you to verification. everyone has a plan (analysis) until they get punched in the face (actually building the thing)
2
6
1,208
We still underestimate how profound this technology will be.
I wouldn’t be surprised if in the coming 5 to 10 years huge hardware incumbents that are more than 100 years old die after losing the ability to compete with significantly smaller emerging hardware companies.
We keep underestimating second order effects of new tech and how much change it can do beyond what is already the status Quo.
The main thing that might stand in the way is the inertia with which societies move. You can’t turn around a tanker on the spot after all.
3
360
We need to start seeing software as a “calculator”, the real value of the calculator is the work done with it and not the fact it can add numbers easily.
claude.com/resources/article…
1
106
Who is buying this data?
If these prices keep going up, at some point it will make sense to design and manufacture hardware just to sell the data
Are you crazy Natalie? No we are not selling our data. Data is what we use to make our robots smart. I am worried small mom and pops shop would fall for this though. Getting one of these almost every day now. Who is behind this?
100
I talk about design automation a lot, here is another one:
With design automation getting more mainstream, we will start seeing more bootstrapped hardware companies, it will start with simple products, maybe even consumer parts, but it will very fast move more towards enterprise and B2B.
But we still need on demand, 48 hour delivery, low volume manufacturing in Europe to do that...
124







