From subwoofers to microfactories. Fusing local AI, physics & deep creativity. Also made a feature film about Star Wars prequels once. Keep it simple.
San Antonio, TX
- Tweets 394
- Following 472
- Followers 147
- Likes 1,266
Pinned Tweet
Ran 7 embedding models on 54 notes. All scored above 90%.
Then ran them on 995 notes. The ranking inverted. The winner fell to #4.
Copying their winner is copying their corpus.
Do your own evals. #RAG #Embeddings #BuildInPublic
Harness + dataset public: @ollama @huggingface
1
1
148
Let the blaspheming begin. @OmarchyLinux @dhh @0xBOYD
#omarchy installed on my 2019 MacBook Pro 13” 8gb.
It was stupid simple. Nothing like the old days.
1
2
104
I wanted a way to leverage my omp harness in other agent harnesses.
The other harnesses could set this up on its own but varied with each use.
So I made this instead: github.com/kylebrodeur/omp-r…
Any feedback is greatly appreciated! #aiagents
1
54
The #knowledge that keeps a shop running often lives in one person. When they leave, the tools remain. The judgment goes with them.
I designed #microfactory as a system of ideas to keep that knowledge from going away.
My first experiment was to see how I could leverage small models and my #3Dprinting hobby with a compounding knowledge and skill system. A system that learns job after job.
Built on @googlegemma
Training pipeline on @modal
GGUF artifacts on @huggingface
1
41
I also kept the bad measurement: 32.6% accuracy across 178 real FDM records. The simulator is right on clean successes and blind to moderate failures. That limitation stays visible.
1
6
The fine-tuned model can run locally through @ollama, so the system has no API key or monthly bill after it is installed.
What matters... Preserve the judgment, expose the uncertainty, and keep a human at the gate.
Read more here kylebrodeur.substack.com/p/m…
If there are any nerds that have found this interesting, please reach out 😃!
14
Why is it why I see "Fireworks" I think of this? #ripmacromedia
GLM-5.3-Flash is live on Fireworks. Day 2.
Why not Day 0?
Because being first isn't the goal. Being correct is.
We unplugged peculiar behaviors of over-thinking from the initial tests, and shared all fixes back to open source.
Quality trumps hype. Let's hold a high quality bar together.
31
Y'all need to stop SLACKING on your photo upload UIs
If I upload a HEIC, AVIF or literally any format. JUST CONVERT IT IN THE BROWSER
It's the image is too large for you, RESIZE IT IN THE BROWSER
If I drag and image anywhere in your application HANDLE IT INSTEAD OF REFRESHING THE PAGE
We have the technology! Unacceptable
140
200
5,212
189,319
ThoughtDAG is pretty impressive.
I've been working on similar projects but haven't gotten that far with a UX that I wanted.
Now I can explore how this was made instead: github.com/chenxiachan/thoug…
19
THIS
I built a TUI design tool for my work at @opencode and I’ve been using it for the last 2 weeks
→ Design directly on the terminal grid
→ ASCII borders that join automatically
→ Flexbox, multiplayer, sections etc
tooey.design
1
57
We're introducing Run SDK: secure 𝚎𝚟𝚊𝚕 for dynamic Code Mode execution.
When agents write code, you don't always need a full sandbox. You can 𝚛𝚞𝚗 their code in a lightweight QuickJS secure context. Faster and more cost-efficient.
𝚗𝚙𝚖 𝚒 𝚛𝚞𝚗
vercel.com/blog/introducing-…
29
An indoor garden gave me a theory: planting is the exciting part. Maintenance is the real work.
So I built the literal version: a sensor rig that watches plants, logs what it sees, and keeps every reading on hardware I own. #OfficeGardener #diy #PlantBased #electronics
2
1
41
Update: the rig's dashboard is public now. You can watch the plant data live at grow.officegardener.com
The brain stays local, the readings stay on my hardware, and the view is open to anyone who wants to look.
6
Most AI assistant work starts with a blank chat.
Explain who you are. Explain your tools. Paste the doc. Describe the process. Ask the question.
Then do it again tomorrow.
The assistant is capable, but it has amnesia. #aiagents
1
1
2
192
Files are for context every conversation needs.
Tools are for live information.
Memory is for stable facts.
Skills are for repeatable workflows.
Signposts are for everything else.
1
3
The assistant is only as good as the operation behind it.
Put #knowledge in the wrong layer and you get stale answers or clogged context.
Put it in the right layer and the tool starts to know how your team works.
Full post:
uof.digital/claude-ask-your-…
4
@VictoriaMetrics on a Pi at home. No cloud subscription.
The rig is running now. This is the setup, not the experiment.
1
4
30