Alexandre Carlhammar (@acarlham) header
Alexandre Carlhammar (@acarlham) profile picture

‘Freak of all trades’ | Building The Final Factory | prev @Stanford @EPFL

Palo Alto, CA
Joined July 2025
Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)Photo by Alexandre Carlhammar (@acarlham)
Thinking about this a lot
1
2
7
397
Banger
You could fold a billion towels and still not get close to building a Dyson sphere
3
310
Alexandre Carlhammar retweeted
Kutztown Foundry, pouring metal in Pennsylvania since 1866, is now Hadrian Factory 6. We are investing $300M to upgrade and expand it, adding the casting capacity America's naval and munitions supply base needs.
59
188
78
1,427
181,842
Alexandre Carlhammar retweeted
Replying to @emm0sh
Yes, the only way this is solved is by looping in the data from deployed assembly line.
1
3
273
We just took #1 on CADGenBench, the leading benchmark for parts engineering. huggingface.co/spaces/Huggin… We developed a SOTA CAD generation harness through a novel predictor-corrector architecture.
20
22
7
245
130,732
The model proposes the part. A deterministic geometry kernel then verifies the B-rep against dimensional and topological constraints that two independent agents extract from the drawing. We also reinforced the system on large libraries of real manufacturing and assembly-line data that we collected. More on it soon.
1
2
20
2,074
My kind of people
1
20
957
Alexandre Carlhammar retweeted
Anthropic Launches Claude for Manufacturing
70
78
35
1,793
211,561
Low-volume, high-mix manufacturing presents 3 fundamental challenges for automation: - Design: Products aren’t designed for automated assembly. - Deployment: Automated assembly lines are bespoke to individual products, with little skill transfer to new products. - Flexibility: Adapting lines to changes in product design, supply variance, etc. is incredibly difficult. We’re fixing this. More on it soon.
This quoted post is unavailable.
8
2
1
99
381,710
Alexandre Carlhammar retweeted
Don't get lost in the sauce. Focus on what you can control: 1. The folks in your corner. 2. Your personal relationships 3. Executing your purpose / mission. Let's rip it. Time to build.
7
4
58
2,140
Alexandre Carlhammar retweeted
Factories of the future will not only be factories As the performance of design automation tools increases, manufacturers will start moving more upstream. Design, simulation and testing will all get compacted into one comp. The manufacturer will design the parts The manufacturer will simulate the part The manufacturer will manufacture the part The manufacturer will test the part Not every manufacturer will be able to profit from this consolidation though.
1
3
1
22
1,775
Alexandre Carlhammar retweeted
"getting it manufactured still required several rounds of review and iteration" yes, welcome to the real world. we could have solved rsi and it won't matter. the only thing that solves this will be factories.
Replying to @OpenAI
Two OpenAI team members worked with GPT-6 Astra to design a prototype device for talking to Codex in under a week, as a side project. Astra designed the enclosure, circuit board, interface, and software. Getting it manufactured still required several rounds of review and iteration.
15
7
223
10,881
Imagine the magic you could create if you also had manufacturing signals in the loop…👀
1
4
757
Alexandre Carlhammar retweeted
You cannot scale manufacturing without systems and scaleable processes. If there's a problem, you have to attack it right away with a solution that is easily repeatable across hundreds of new hires. Very hard to do. But if you do it, you will win.
13
17
5
271
9,350
Alexandre Carlhammar retweeted
Today, we are introducing Fern: our first model that supports up to 10M tokens in context window!🌿 You could put an entire repository (or repositories!) in the context, continue to work on problems for days without hitting compaction limits or use it to consolidate other agents’ works. 
 As always, we have a lot more work to do but wanted to give you the update. Excited hear how you would use a 10M token context window.
Introducing Fern, Mirendil’s first model: pretrained from scratch, with up to 10M tokens of context! That’s ~0.5M lines of code (roughly Megatron-LM scale) or days of an agent’s work, all in context at once. Its first version, Fern 0.1, had a short round of post-training 🚢
11
13
1
173
8,402
Alexandre Carlhammar retweeted
9
527
22
3,414
50,259
From thinking of a product, captured through BCI, to generatively designed products (CAD, PCB) and co-designed assembly lines, all within a shared agentic engineering and manufacturing harness (FEA, CFD, DFM, …). Bidirectional signals across the entire system, continuously co-optimizing product and factory for cost, throughput and volume. Driving the distance between thought and physical reality to zero. Soon, this won’t be science fiction.
2
8
5,466
Alexandre Carlhammar retweeted
Hey anduril
We made this omnidirectional "swerve" robot that's about 2.5 inches thick. It's holonomic, meaning it can move in any direction, spin in any direction, simultaneously. There is a quick swap turret with slip ring that gives it even more agility. planning on an especially fun attachment soon, but you can add cameras, lidar, robotic arms, lasers, to name a few. Files are free on my ko-fi. Mechanical BOM is just $110 Even with electronics it's well below $200. For comparison, most robotics competitions use swerve modules (this one has four such modules), and each individual module costs about $500-$1000. Part of the reason i hate competition robotics Swerve is in no way new but what about a thin slab one? Wondering if there are any actual use cases of this thing. Built with @JamesMakes29, Thank you @Willqyu at The Standard Manufacturing Company for helping support Additionally, thank you LDO motors, SendCutSend, PCBWay.
35
44
11
1,117
36,974
Alexandre Carlhammar retweeted
There is a missing link here. Modern manufacturing is built on programmable, generalisable machines. CNC mills, lathes, lasers, and more. One machine can make many parts. 1/5
the evolution of production is pretty clear: solve design/quoting and have it connected to a factory, immediate margin visibility, fabrication strategies and sequencing to reinforce zero rework/oneshotting every job. extremely scalable, multiples on these will skyrocket.
7
7
80
118,510
Alexandre Carlhammar retweeted
the reality is (a) true reindustrialization requires you to completely rebuild production methods from scratch with new tech, (b) part (a) takes a shitload of r&d, (c) only once you've proven a better method should you scale it, (d) these companies are far more valuable than PE multiples, (e) taking over someone's machines for revenue is fine but isn't really solving anything
Pre-seed valuations in reindustrialization are climbing, even before prototypes or production. At these valuations, why not take a PE approach after the raise? Acquire machine shops from retiring owners. Get revenue, skilled labor, equipment, and production capacity on day one. Build your new products on the night shift while expanding capacity. When equipment lead times are a major bottleneck, buying existing capacity could be a massive advantage. I think we’re about to see a new monster emerge. A hybrid of venture capital and private equity built for American reindustrialization.
17
15
2
208
8,956