Add generative-CAD outreach-contacts background note backed by Edison query

[Copilot]

Resolves the request to identify named individuals and organizations the lab can contact for help with generative CAD, with verifiable public contact channels for each.

Follows the Edison-driven paper/background/ pattern from #29 (see also #7, #27).

• Runner — paper/background/edison_run_outreach_contacts.py. Single LITERATURE_HIGH task with the prompt embedded verbatim. Prompt explicitly forbids fabricating emails and requires a public source URL per entry; covers 7 categories (academic CAD-LLM, code-CAD OSS, commercial CAD-API, SDL groups, AM experts, communities, funding).
• Raw artifacts — paper/background/edison_artifacts/gencad_outreach_contacts.{task.json,answer.md,references.md}. status=success, ~42 k-char answer, 11 references with DOIs.
• Synthesis — paper/background/06-generative-cad-outreach-contacts.md. ~45 named contacts organized into 7 categories with markdown subsections (### Name) for each entry, making each contact easy to link to directly. Quick-reference table at the top, then per-category subsections. Every entry ends with (source: <full URL>) pointing at the page where the listed channel (email / GitHub / X / lab page / Discord invite) can be verified. Where no public personal email exists (e.g., Faez Ahmed, Aspuru-Guzik, Nick Kallen), the entry says so and falls back to a lab/group/social channel rather than guessing.

Coverage highlights: DeepCAD (Wu, Zheng @ Columbia), Text-to-CadQuery (Xie, Ju @ ASU), CAD-Recode (Rukhovich, Aouada @ uni.lu), CADSmith (Barkley, Farimani @ CMU), ArtiCAD (Yu @ Beihang), SketchGraphs (Seff, Adams @ Princeton), Fusion 360 Gallery (Willis @ Autodesk), CAD-Coder / GenCAD (Doris, Ahmed @ MIT), Sadik @ Honda RI; CadQuery / build123d / OpenSCAD / FreeCAD maintainers; Zoo.dev/KittyCAD (Frazelle), McNeel (Baer), Plasticity (Kallen); Acceleration Consortium, Cooper, Cronin, Abolhasani, Schrier, Brown; Simpson, Beese, Pollock, Vecchio; CadQuery/build123d Discords, FreeCAD forum, IDETC-CIE, SCF, CAD'25; NSF DMREF / Future Manufacturing, ARPA-E DIFFERENTIATE, America Makes.

New files only — no overlap with #29's runner or README.md, so the two branches can land in either order without conflict. The synthesis README row in paper/background/README.md (introduced by #29) is intentionally not touched here; add it in a follow-up after #29 merges.

[sgbaird]

@copilot split up the authors into markdown sub(sub?)sections so it's easy to link to specific entries

[Copilot]

@copilot split up the authors into markdown sub(sub?)sections so it's easy to link to specific entries

Done! Converted all bullet-point entries across all 7 sections into markdown subsections (### Name). Each person, organization, community, and funding program now has its own anchor for easy linking — e.g., #rundi-wu, #cadquery-discord, #nsf-dmref. All content, citations, and source URLs remain unchanged. Commit: 89648e7

[sgbaird]

Shortlist for @swcharles and @williamulbz

• https://github.com/vertical-cloud-lab/powder-doser/blob/copilot/determine-helpful-contacts/paper/background/06-generative-cad-outreach-contacts.md#jesse-barkley (make sure to cc https://github.com/vertical-cloud-lab/powder-doser/blob/copilot/determine-helpful-contacts/paper/background/06-generative-cad-outreach-contacts.md#amir-barati-farimani) - let's ask about having a conversation to get feedback on how we've been leveraging CADSmith so far
• https://github.com/vertical-cloud-lab/powder-doser/blob/copilot/determine-helpful-contacts/paper/background/06-generative-cad-outreach-contacts.md#adam-urbanczyk ; let's ask to have a conversation about recommendations on using LLMs with CADQuery and see if we can get some feedback on our current usage
• https://github.com/vertical-cloud-lab/powder-doser/blob/copilot/determine-helpful-contacts/paper/background/06-generative-cad-outreach-contacts.md#jessie-frazelle - let's reach out asking about potential academic research collaboration around design for automated metal 3D printing (e.g., powder dosing). We can try the email in https://verticalcloudlabbyu.slack.com/archives/C0B3CGGV4VC/p1779293517201589?thread_ts=1778701701.547399&cid=C0B3CGGV4VC; if that bounces, let me know

Please make sure to CC me as well, and if Sam sends an email, cc Will too, and vice versa. Run the draft by me via slack first before sending

[sgbaird]

Just sent a linkedin message to Jess:

Would love to chat about a possible academic collaboration, leveraging Zoo tools for metal 3D printing projects (#7 (comment))

[sgbaird]

@swcharles @williamulbz @lbwinters how is outreach going?

[sgbaird]

Please update with who has been contacted already

[sgbaird]

Sorry to pester 🫣

[sgbaird]

@claude add draft emails here in your comment reply for #43 (comment)

[sgbaird]

Lol.. this is pretty bad. I'm fully at BYU, not at the AC anymore. I'm going to hide the above response to avoid clutter. @claude, create new drafts. Look to vertical-cloud-lab/tensegrity-optimization#41 for some pointers and ideas. Include specifics, none of these placeholders. Luke will be sending the emails. You can leave those drafts in a committed file, but make sure to include the direct link in your comment reply

[claude]

Claude finished @sgbaird's task in 4m 12s —— View job

---

New outreach email drafts committed

• Gather context (shortlist, contacts file; corrections: Sterling fully at BYU, Luke Winters sending)
• Review tensegrity-optimization#41 pointers (artifact-grounded pitches, concrete single asks)
• Mine repo for specifics (CADSmith runs, CadQuery assembly, Zoo API/ZDS usage)
• Write draft emails to a committed file (no placeholders)
• Commit + push, link the file here

📄 Drafts: outreach/generative-cad-email-drafts.md (commit ee5ddb2)

Three drafts, written for Luke Winters as sender (cc Sterling on all; BYU affiliation only — no AC mention). Each is grounded in what the lab has actually done, with repo links the recipients can open:

1. Jesse Barkley, cc Amir Barati Farimani — feedback call on our CADSmith usage. Cites the real runs from PR #59: both printable plates converged on iteration 0 (3 LLM calls, 45–67 s, volume within 0.5–1.1 % of hand-authored references), the three patches we made to get it running (offered as upstream contributions), the single-part-vs-assembly limitation we hit, and the PR #105 three-tool comparison prompt.
2. Adam Urbanczyk — conversation on LLM-generated CadQuery. Cites the PR #112 interference-gated parametric assembly (params.py, gear/screw/thread helpers, 10-rule interference_report()) and OCP CAD Viewer adoption (PR #70), with three concrete questions. No public email exists, so the routing notes say to send it via CadQuery Discord DM or a GitHub Discussion — the text works in either channel.
3. Jessie Frazelle — academic collaboration on design for automated metal 3D printing. Acknowledges Sterling's 20 May LinkedIn message, then leads with paid usage: all 11 parts generated via the Text-to-CAD API with KCL + STEP + spec-verified bbox manifests (PR #110), the ZDS unlimited subscription + best-practices note (PR #93), hand-edited KCL (PR #66), and proposes specific collaboration angles (Text-to-CAD benchmarking on manufacturability-constrained parts, KCL as an agent target, Digital Discovery case study, NASA Space Grant tie-in).

The file opens with routing notes for Luke: cc rules, run-by-Sterling-on-Slack-first, re-verify addresses at their source URLs, and that Jessie's address comes from the lab Slack thread (with the link) — if it bounces, tell Sterling.

The one thing I could not include: Sterling's BYU email address and Jessie's actual address aren't public, so the Cc/To lines name them without an address rather than inventing one — Luke fills those in when sending.
· Branch

[sgbaird]

@claude these need to be much shorter. Keep it brief. Send to edison scientific for review. Implement feedback. We don't need to have a lot of specifics in here, but at least an acknowledgement of their work and its relevance and enough to go on for some of the things we've been doing on our side with generative CAD.

[claude]

Claude finished @sgbaird's task in 6m 12s —— View job

---

Drafts shortened, Edison-reviewed, and revised

• Read current drafts and Edison runner pattern
• Rewrite the three drafts much shorter (~120 words each, down from ~250–300)
• Send shortened drafts to Edison Scientific for review (ANALYSIS task, status=success)
• Implement Edison's feedback
• Commit artifacts + revised drafts, push (commit d88276e)

📄 Revised drafts: outreach/generative-cad-email-drafts.md

Each email is now three short paragraphs: one line of who-we-are, one acknowledgement of the recipient's work with a single credibility signal (one repo/PR link), and one clear ask.

Edison review (full review, runner, task.json) — verdict was "minor edits" on all three. Changes applied:

• Barkley: named the metric ("within ~1 % … by volume"), collapsed to a single ask ("compare notes on our CADSmith results and assembly-level generation" instead of the one-sided "give us feedback"), cut the patches-to-upstream sentence, subject → "CADSmith on BYU powder-doser parts".
• Urbanczyk: "most of our CadQuery code is LLM-generated" (not "most of the code"), reworded the extractive "could we get your advice", cut the contribute-back offer, and added Edison's channel guidance to the routing notes — Discord DM preferred with a short sign-off; if a public GitHub Discussion, drop the call request and ask asynchronously.
• Frazelle: removed the paying-subscriber line and the paper-case-study mention (both flagged as transactional/premature), replaced "load-bearing" with a plain factual sentence about the eleven Text-to-CAD parts, softened to "see whether there is a useful fit for an academic collaboration", subject → "Zoo Text-to-CAD for BYU metal-printing research", and the May LinkedIn note is mentioned once without pressing for a reply.

Routing notes at the top of the file are unchanged in substance: Luke sends, Sterling cc'd on all, run past Sterling on Slack first, Jessie's address from the lab Slack thread.
· Branch

[lbwinters]

@claude @sgbaird In order to dm Adam Urbanczyk, I will need his exact discord username or to share a server with him. I couldn't find a linked server in the cadquery github page. Claude can you check on that and see if you can find that? Additionally, I'm curious if there's a way we could connect with him through github or linkedin?

[claude]

Claude finished @lbwinters's task in 2m 51s —— View job

---

Adam Urbanczyk contact channels — found and verified

• Find the CadQuery Discord server invite link
• Try to find Adam's Discord username
• Check GitHub contact options (profile, Discussions)
• Check for a LinkedIn profile
• Update the routing notes in the drafts file with findings (commit 46fa59a)

@lbwinters here's what I found:

Discord — server found, invite verified

The reason you couldn't find it: the server isn't called "CadQuery". It's named "Python OCP CAD" — a shared community server for CadQuery, build123d, and the other OCP-based Python CAD tools, and the invite drops you into its #cadquery channel.

Invite: https://discord.com/invite/Bj9AQPsCfx — I verified it today via Discord's public invite API: it's valid, permanent (expires_at: null), ~1,580 members. It was posted by CadQuery co-maintainer Jeremy Wright in CadQuery/cadquery#1680 (an issue about older invites appearing broken — that turned out to be a browser problem, and this link was confirmed working).

Adam's exact Discord username is not published anywhere I could find (GitHub, Google Groups, FOSDEM pages). Once you join, search the member list for "adam" / "adam-urbanczyk". If he's not there or doesn't respond, don't wait — use GitHub, which is his demonstrably active channel (see below).

GitHub — the most reliable channel

• He's @adam-urbanczyk, and he's currently active: his public activity shows CadQuery PR reviews and pushes to CadQuery/OCP as recently as 19 Jul 2026 (four days ago).
• His profile shows no public email, so a direct email via GitHub isn't possible. The practical route is a GitHub Discussion on CadQuery/cadquery — this is also what the Edison review recommended as the alternate channel, using the async version of draft Initial design diagram, brainstorming, and open parametric CAD pipeline for powder-excavator #2 (drop the call request, ask the technical questions so the community can weigh in).
• One more option from the CadQuery README: info@cadquery.org reaches the core maintainers by email. It's framed for consulting inquiries, but it's a legitimate published address if you specifically want an email thread.

LinkedIn — profile found

His profile is linkedin.com/in/adam-urbańczyk-82511251 (Adam Urbańczyk, ASML, Netherlands). I'm confident it's him: his FOSDEM speaker bio says he's an applied physicist with a PhD from TU Eindhoven living in Utrecht, working as an algorithms engineer for "a large semiconductor equipment vendor from Holland" — i.e., ASML. A connection request with a short note works as a fallback, but he's clearly most reachable where he already spends time: GitHub, and possibly the Discord.

Suggested order

1. Join the Python OCP CAD Discord via the invite above, look him up, and DM the draft if he's there.
2. Otherwise post the async version as a GitHub Discussion (he actively watches the repo).
3. LinkedIn or info@cadquery.org as fallbacks.

All of this is now recorded in the routing notes of outreach/generative-cad-email-drafts.md (commit 46fa59a), so it travels with the drafts. Per the existing routing notes, still run the message past Sterling on Slack before sending.

--- · Branch

[lbwinters]

@sgbaird here are the current email drafts. If you have better subjects you want included, you can put them here too, and I'll change the email drafts:

First, to Jessie Frazelle:

Hi Jessie,
I'm Luke Winters from Dr. Sterling Baird's Vertical Cloud Lab at BYU — Sterling mentioned this in a short LinkedIn note in May. We build an open-source powder doser for automated metal 3D printing (https://github.com/vertical-cloud-lab/powder-doser), and we used Zoo's Text-to-CAD API to generate KCL and STEP files for all eleven printed parts (#110).
We'd like to see whether there is a useful fit for an academic collaboration on design for automated metal 3D printing. One possible project is benchmarking Text-to-CAD on manufacturability-constrained parts; we have specifications, reference CAD, and printed hardware for evaluation.
Would you be open to a 30-minute call in the next few weeks?
Best, Luke Winters Vertical Cloud Lab (PI: Dr. Sterling Baird), Brigham Young University

Then, the one to Jesse Barkley:

Hi Jesse,
I'm Luke Winters, a researcher in Dr. Sterling Baird's Vertical Cloud Lab at BYU (cc'd). We build an open-source powder doser for metal-AM workflows (https://github.com/vertical-cloud-lab/powder-doser), generating most of our CAD with LLM pipelines — which led us to CADSmith.
We've run CADSmith on real parts: two of our printed plates converged on the first iteration, within ~1 % of our hand-authored CadQuery references by volume (#59).
Would you and Prof. Farimani be open to a 30-minute call to compare notes on our CADSmith results and assembly-level generation?
Best, Luke Winters Vertical Cloud Lab (PI: Dr. Sterling Baird), Brigham Young University

[lbwinters]

@sgbaird also here is the draft to adam urbanczyk:

Hi Adam,

I'm Luke Winters, a researcher in Dr. Sterling Baird's Vertical Cloud Lab at Brigham Young University. We're building an open-source, auger-fed powder doser for metal-AM powder handling (https://github.com/vertical-cloud-lab/powder-doser), and CadQuery is our primary CAD medium — with the twist that most of the code is LLM-generated and then human-reviewed.

Concretely, our full parametric assembly lives in CadQuery: a params.py single source of truth for every dimension, helpers for involute spur gears/rings, helical Archimedes screws, and printable single-start threads, and an interference_report() that gates the build on ten spec rules — bore openness, gear mesh without jam, running fits, and one deliberate 3 mm solenoid interference (#112). Individual parts (split shaft-collar brackets, a tap collar, hinged mounting plates) were authored the same way, and we've adopted OCP CAD Viewer as our local authoring loop (#70).

We'd love a conversation — a short call or async here on Discord, whichever you prefer — about:

Your recommendations for making LLM-written CadQuery robust: is there an API subset or idiom set you'd steer code generators toward, and which OCCT failure modes do you see people hit most?
Feedback on our current usage (the interference-gated assembly above) — anything that looks fragile or unidiomatic to you?
Whether the failure/success patterns we've logged from LLM-generated CadQuery would be useful to the project, e.g. as docs or examples we could contribute back.
Thanks for maintaining CadQuery — it's been the backbone of this project.

Best, Luke Winters Vertical Cloud Lab (PI: Dr. Sterling Baird), Brigham Young University https://github.com/vertical-cloud-lab/powder-doser

[sgbaird]

@lbwinters please use the following and include a link to one of our YouTube videos along with a picture of the powder doser

Jessie Frazelle

Hi Jessie,
I'm Luke Winters from Dr. Sterling Baird's Vertical Cloud Lab at BYU. We built an open-source powder doser for automated metal 3D printing, and we used Zoo's Text-to-CAD API to generate KCL and STEP files for various parts (#110).
We'd like to see whether there is a useful fit for an academic collaboration and also get feedback on optimal use of Zoo. One possible project is benchmarking Text-to-CAD on manufacturability-constrained parts.
Luke Winters
Vertical Cloud Lab @ BYU

Jesse Barkley

Hi Jesse,
I'm Luke Winters, a researcher in Dr. Sterling Baird's Vertical Cloud Lab at BYU. We built an open-source powder doser for workflows involving 3D printable metal powders, generating most of our CAD with LLM pipelines. Would you be open to chatting about best practices and feedback on some of our approaches with generative CAD?
Luke Winters
Vertical Cloud Lab @ BYU

Adam Urbanczyk

Hi Adam,

I'm Luke Winters, a researcher in Dr. Sterling Baird's Vertical Cloud Lab at Brigham Young University. We built an open-source powder doser for workflows involving 3D-printable metal powders, primarily using generative CAD. Would you be willing to chat and help provide some perspectives on generative CAD?
Luke Winters
Vertical Cloud Lab @ BYU

[lbwinters]

Outreach has been sent, I will update with responses!

[lbwinters]

@sgbaird Jessie Frazelle responded with the following:

"Sure would you like some free credits? What email is your account under?"

How would you like me to respond?

[sgbaird]

@lbwinters could you create a zoo account with the VCL Gmail and then send that one asking for the credits and saying that that would be great ?

[lbwinters]

@sgbaird Yes! Where can I find the login credentials for the VCL email? I haven't signed in before

[lbwinters]

@sgbaird whoops never mind I just saw your message