←
Steve174 posts
Steve avatar

Steve

@scfred

Engineer. Mostly buildings, lately software.

◷ Joined July 2015▧ 32 Photos and videos
Steve avatar
Steve@scfred·5m

Coding agents can draft CAD sheets now. tsdraft is a TypeScript library for it: the drawing is code, the output is DXF & PDF, every safe redraws the sheets. Here an agent edits a 15-sheet permit set, and tsdraft flags it when the house lands 5ft into the front setback.

0:30
Steve avatar
Steve@scfred·2m

Agents drafted this 15-sheet permit set for a house with it, every sheet drawn from one model of the house.

Post attachment
Steve avatar
Steve@scfred·2m

Scripting CAD today: - COM or a CAD app's console: Windows, a licensed seat, & a script that stops mid-command hangs till it times out - ezdxf: writes a layer named "X|Y" or "WALL ", which the leading CAD app refuses tsdraft runs anywhere Node does.

Post attachment
Steve avatar
Steve@scfred·Nov 12

I've been playing with @theo's tstack skills on my new omarchy machine in @t3dotcodes. Was curious how the council reviews were going, so I asked. T3Code rendered all of this inline, then hucked it into an @executorsh artifact for good measure. You just ask for things and the agents make it so. Looking forward to an improved Grok 4.8...

Post attachmentPost attachmentPost attachmentPost attachment
Steve avatar
Steve@scfred·Nov 2

Thanks for the shout out, looking forward to wrapping up the hackathon! Will be publishing a full rundown on the final product and future plans! @MetaforDevs Ended up going further w/ SAM 3.1 too, using it for rough highlighting of review comments!

🧵 (17/22) Construction plans by @scfred: Muse proposes candidate fixtures from the plans, SAM 3.1 masks the plumbing fixtures, and a review pass matches them against the lavatory and WC schedule.

Roman Rädle avatar
⟳ Steve retweeted
Roman Rädle@raedle·Nov 2

🧵 (17/22) Construction plans by @scfred: Muse proposes candidate fixtures from the plans, SAM 3.1 masks the plumbing fixtures, and a review pass matches them against the lavatory and WC schedule.

Steve@scfredOct 27

plumbing segmentation is working pretty good for the meta hackathon. muse proposes candidates from the plans, sam 3.1 does the masks, and the review matches the lav/wc schedule so far. moving onto electrical next then structural and civil @MetaforDevs

0:08
Steve avatar
Steve@scfred·Oct 29

I know Claude is already doing this, but it'll be so nice to be able to play around with the decisions API and share small internal apps without having to setup billing, etc w/ these credits

Starting today, eligible ChatGPT Pro and Business subscribers can claim up to $500 per month in API credits to build apps and agents.

2:25
Steve avatar
Steve@scfred·Oct 27

plumbing segmentation is working pretty good for the meta hackathon. muse proposes candidates from the plans, sam 3.1 does the masks, and the review matches the lav/wc schedule so far. moving onto electrical next then structural and civil @MetaforDevs

0:08
Steve avatar
Steve@scfred·Oct 21

Doing my first hackathon, got approval yesterday! Super pumped to build!

Build the future of superintelligence with us. The Meta Global AI Developer Hackathon is taking place from Oct 26-Nov 4. Apply to compete for a share of $1M in prizes and build with our newest models. Each selected participant gets $150 in Meta Model API credits. Limited to 20,000 selected applicants. Terms apply → bit.ly/4saQrA8 Apply now → bit.ly/4hR3cuQ

0:10
Steve avatar
Steve@scfred·Oct 14

If anyone is still miffed w/ openai re: usage limits being cut in half, hopefully this helps explain Usage from the last 3 months on @t3dotcodes 6.89B 5.6 Sol tokens vs 7.59B 6.1 Sol tokens GPT-6.1-sol cost 38% as much as GPT-5.6-sol, despite having 10% more usage. Yeah having usage dollars cut sucks, but those dollars go much farther w/ the new model Now Astra... Astra is a different story

Post attachment
Steve avatar
Steve@scfred·Oct 14

parsing through a pile of construction docs working on a benchmark, deepseek hit me with the "hold up, wait a minute..."

Post attachment
Steve avatar
Steve@scfred·Oct 13

As a follow up, all 24k sorted in about 2 hrs w/ grokbot babysitting everything, ~$15 total the bliss of inbox zero :) now on to automating unsubscribing from years of newsletters and other crap

Steve avatar
Steve@scfred·Oct 13

working on migrating icloud & outlook into my gmail account and inherited 14k emails going back years... not sorting that by hand lol so I had claude w/ executor v2 build a gmail style inbox that sorts them for me. clef runs most of it on cloudflare's worker paid tier, and jev from @typesafeai takes the ones clef is unsure about. the jev hype may have died down, but together they're getting through the pile for pennies 550/14k sorted so far for about $0.52 total. each one gets keep, archive or delete plus a confidence level, and nothing changes until I approve it. all built w/ @executorsh

Post attachment
Steve avatar
Steve@scfred·Oct 13

it'd be kinda sick if the new Decisions API worked w/ Codex sign-in, not just API keys. cheap fast routing inside personal software would be a great fit @thsottiaux