PartFoundry

Day 3 of 30 — game-day switch

This morning’s plan was Florida Condo Tracker. By 9am Sunday I’d looked at the FCT branch, didn’t like the state of repair, and pulled a fresh build off the bench instead: PartFoundry — an on-demand replacement-parts manufacturing platform serving the $15B+ obsolete and discontinued parts market.

One quota session. 9:45am scaffold to ~12:10pm quota-reached banner. Three commits. Live at partfoundry.shop, repo public at github.com/toddamerrill/part-foundry, business plan published as a sibling doc.

This is also the first fully public build of the 30-day run. Repo open, business plan open, screenshots of the actual quota session embedded below. If you want to see what one Sunday morning of agentic coding actually looks like — start to quota — it’s all here.

What it does

The original manufacturer doesn’t make it. The aftermarket gave up. The forum thread ends with someone offering to trade a kidney. That’s where PartFoundry starts.

Three service tiers, one platform:

• Catalog ($25-75) — pre-modeled parts already in the library. Vintage car knobs, appliance shelf clips, industrial housings. Order, print, ship. 3-7 days.
• Scan & Print ($75-150) — customer ships the part, we scan it, we print it. The reference part comes back with the replacement.
• Reverse Engineer ($150-500) — broken or partial part. Materials science + CAD inference + symmetry reconstruction. Taylor’s plastics-engineering call decides the material; Aaron’s CAD work renders it; the press-fit lands on the bench.
• AI Scan / PartScan (Phase 3, provisional patent track) — smartphone-only capture with AR-guided quality feedback and AI geometry completion. Months 9-18 on the roadmap.

Materials in stock today: PLA, PETG, ABS, Nylon (PA12), Resin (Tough 1500), and metal via partner network (aluminum / stainless / brass).

Founding team

• Taylor Merrill — Founder & Chief Engineer. Vintage Car Addict, Plastics Engineering pioneer, Rubbermaid production background. Owns material selection and quality.
• Aaron Vedlitz — Production Lead & Design Partner. CAD, 3D modeling, post-processing. Aaron is also @eldritchcosplays — an award-winning prop fabricator (Best Craftsmanship at DragonCon, Best in Show at MomoCon) whose Instagram archive is the proof on the landing page. Same hands, same standards, applied to your part.

The Eldritch Cosplays archive is the on-ramp. Cosplay is not the market — it’s the credential. A craft community that understands what a one-off, dimensionally accurate, finish-quality fabrication actually costs is the easiest first audience to talk to about $25-500 replacement parts. The thesis spreads outward from there: vintage automotive (Model A Ford Club, Vintage Mustang Owners, Porsche 356 Registry), appliance repair (Park Appliance and the long tail of independent technicians), and industrial maintenance (Reed Industrial, Haas spindle work, Bridgeport mill parts).

Why Day 3, why a game-day switch

The original Day 3 was Florida Condo Tracker — a regulated-compliance autopilot for HB 1203 (post-Surfside structural inspections). I’d planned it as the Sequoia / Julien Bek autopilot-thesis activation slot: sell the work, not the tool. That thesis still applies and FCT will land later in the run, but this morning it wasn’t ready and PartFoundry was. Discipline says ship the build that’s actually ready, not the build that fits the narrative.

The Sequoia activation moves to Day 9 — CogleGroup GP launch, where it lands harder anyway: the partnership announcement is a structural fit for the autopilot frame, and Julien Bek’s thread reply will come off that day.

PartFoundry’s own thesis is different and worth saying out loud: physical goods are a real Velocity Process target. Most of the 30-day run is software ventures. PartFoundry proves the same operating model — agentic coding, single quota session to MVP, public repo + public business plan, design partner / co-founder structure — works when there’s a printer in the loop and physical fulfillment on the back end. The platform shipped today; the printers are ready in the shop; Aaron’s CAD pipeline has been running for years on Eldritch Cosplays builds. The software was the missing piece.

Build session log — 9:45am to quota reached

Three commits, one quota session, one Sunday morning. The screenshots that bookend this section are checked into the repo at the root: Day0_buildthis.png (the prompt) and OneQuotaSessionSundayMorning.png (the quota-reached banner).

09:45 ET — f177ca8 Initial scaffold: Next.js 14 + Cognito + DynamoDB + S3 + ECS Express CDK. Read CLAUDE.md, the Technical Product Design doc, and the Business Case doc. Scaffolded the Next.js 14 frontend (App Router, standalone output, shadcn/ui orange theme), wired AWS Cognito User Pool + Web Client via aws-amplify/aws-jwt-verify, generated the DynamoDB single-table schema (partfoundry with GSI1 + GSI2), set up the S3 presigned PUT/GET flow for photo uploads, sketched the order intake form for all three tiers, and laid down the CDK v2 stacks (data-stack.ts for DynamoDB + Cognito + S3; frontend-stack.ts for ECR + IAM roles for ECS Express Mode). Stripe Checkout + webhook came in alongside the order acceptance flow.

10:45 ET — 46f87b8 Workshop-editorial redesign + sandbox-ready CDK split. The first-pass landing page was generic-SaaS. Replaced it with a workshop-editorial layout: hero (“The part you can’t find anywhere — made and shipped this week”), engineering-callout strip (turnaround / tolerance / starting price / materials count), live-jobs marquee (real entries — 1949 PORSCHE 356 · CARB FLOAT BOWL, 1962 SUB-ZERO 522 · FREEZER SHELF CLIP, 1976 BRIDGEPORT MILL · QUILL FEED HANDLE, etc.), three-tier service tile section, materials grid, three customer testimonials drawn from the GTM target communities. CDK split so a sandbox account can synth/deploy without touching production.

~12:10 ET — quota reached. Mid-Instagram-embed plumbing for the @eldritchcosplays gallery and mid-curl test of the page render. The screenshot is real:

13:12 ET — c958af0 Plumb @eldritchcosplays portfolio + HTTPS at partfoundry.shop. Finished the Instagram embed component (InstagramCard + InstagramEmbedScript — official IG blockquote when given a permalink, fallback image + caption when not), wired the four-slot § 03 — From the Bench gallery in page.tsx, and finished the HTTPS / DNS setup so partfoundry.shop resolves end-to-end. Aaron’s actual permalinks fill in this week as he selects them.

That’s the whole session. Three commits, one quota, one live store. The repo at github.com/toddamerrill/part-foundry has everything — schema, CDK stacks, page source, and both screenshots at the root.

What Claude Code handled vs human judgement

What Claude Code handled: the Next.js 14 scaffold and shadcn/ui orange theme; the AWS Cognito + DynamoDB single-table + S3 + Stripe wiring; the CDK v2 stack split with VelocityStack tagging; the order intake form across all three tiers; the photo uploader with drag-and-drop + presigned PUT; the workshop-editorial landing redesign (hero, marquee, statement, three-tier, materials, testimonials, gallery, CTA); the Instagram embed component with permalink-or-fallback switching.

What required human judgement: the call to publicize the repo and the business plan instead of keeping them private (a Velocity Process bet — public proof scales the inbound that closes the channel partners faster than gated content ever does); the choice to lead the landing page with Aaron’s Eldritch Cosplays archive rather than vintage-car or appliance imagery (cosplay is the most legible craft community for someone arriving cold — once they trust the fabrication chops, the vintage-car and appliance segments are easy to expand into); and the decision to swap FCT off the day-3 slot at 9am instead of forcing through a build whose state of repair I didn’t trust (discipline beats narrative).

What broke: the first version of the Instagram embed loaded embed.js unconditionally — meaning a third-party script ran on every visit even when no permalinks were set. Fixed in the same session: the script now lazy-loads only when at least one card has a real permalink, and cards without one render the local fallback image + caption + Instagram CTA chip.

The numbers

From the public business plan (canonical copy on GitHub; working copy mirrored at builds/PartFoundry-business-plan.md in the VelocityLaunch repo):

• TAM: $15.2B combined (Classic Car Restoration $8.2B / 6.8% CAGR + US Appliance Repair Parts $7.0B / 2.5% CAGR)
• Year 1 target: 1,800 orders × $85 avg = $153K revenue, break-even Month 5
• Year 3 target: 10,000 orders × $105 avg = $1.05M revenue, 34% net margin
• Initial capital: $45K (printers, scanner, materials, working capital, marketing, legal). Self-funded from founder resources.
• Tier 1 unit economics example (vintage car dashboard knob): COGS $17 / sale price $45 / 62% gross profit

The plan is on GitHub. Pull it apart, push back, tell us what’s wrong. Public build means public scrutiny invited.

What’s next this week

• Day 4 (Mon May 4): CompliancePulse — back to the CyberSavi cluster. Already in revenue.
• Day 5 (Tue May 5): CyberSavi Academy — training layer.
• Day 6 (Wed May 6): CyberSavIQ — personal/reputation layer.
• Day 7 (Thu May 7): Theme essay — EU AI Act 2026.
• Day 8 (Fri May 8): GovernAI — closes the cluster.
• Day 9 (Sat May 9): CogleGroup GP partnership announcement — Sequoia / Julien Bek autopilot-thesis activation moves here.

Want to talk

If you restore vintage cars, repair appliances anyone forgot about, or maintain industrial equipment whose original parts catalog has been out of print since the Carter administration — book 30 minutes. Founding-customer pricing (20% lifetime discount) on the first 100 orders.

If you run a vintage car club, a restoration shop, or an appliance repair network and want to talk wholesale or referral terms — same link.

If you’re an operator who wants to see what the Sunday-morning quota session looked like — github.com/toddamerrill/part-foundry.