Safe to wear, able to be remade.
REWEAR-FUSED closes the textile-to-textile recycling loop on children's stretch apparel. It co-designs two matched molecules at once: a new, enzymatically-cleavable elastane fiber, and the matched enzyme designed to cleave it. Around that loop sits a deployable Digital Recyclability Passport that proves each garment is safe and routes it to the right end-of-life.
Built for the Cox Enterprises "Play With Purpose" Sustainability Hackathon, Carter's Making & Remaking track.
-3fb950.svg){kind=icon}
Everything REWEAR-FUSED produces is an in-silico design with benchmark-referenced metrics, not a wet-lab-validated material or enzyme. Every number is a live pipeline output (Engine C), a verified citation, or a clearly-labeled in-silico design value (Engines A/B). No synthetic data, no fictional persona. The honesty is on a
/judgespage that tiers exactly what is real.
Carter's already collects worn baby clothes through KIDCYCLE, then shreds them into insulation and pet bedding, because two walls stop a onesie from ever becoming a onesie again.
- Wall 1, chemical safety. Nobody can prove the recovered fabric is chemically safe for infant skin.
- Wall 2, elastane. Nobody can break down the elastane woven into every stretch garment.
Elastane is in about 80% of all clothing, and as little as 1% of it makes a textile recycler reject the whole garment. Only about 1% of garments are recycled fiber-to-fiber today. REWEAR-FUSED knocks down both walls.
A worn onesie enters. The passport proves it is safe and routes it. We design a new cleavable elastane, design a matched enzyme against its exact bond, the enzyme cleaves it to recoverable monomers, and it re-spins into a new onesie. No one has co-designed a cleavable fiber and its matched enzyme, together, for textiles before.
The closed loop, scrubbed by scroll, at the carbamate-cleavage moment where the two accents fuse.
- Engine A, Cleavable Elastane Design. Constraint-guided virtual screening of polyurethane-urea chemical space. Aliphatic isocyanates only (so nothing carcinogenic comes off near a baby), the cleavable bond in the amorphous soft segment (so the stretch survives), 25 to 35 percent hard segment. It emits the carbamate microenvironment as the enzyme's substrate target.
- Engine B, Carbamate Hydrolase (LigandMPNN redesign). The matched enzyme, designed to cleave the exact bond in Engine A's fiber, with a near-neutral Ser-His-Asp triad so it does not damage companion cotton. Engine B specifies three LigandMPNN redesigns of real serine-hydrolase scaffolds (IsPETase, FAST-PETase, CalB), targeting FoldSeek TM about 0.8 to template and the Lauko PLACER thresholds; the structure shown and downloadable is the CALB / PDB 1TCA reference scaffold the lead design is built on. The redesign and the de-novo fold (no prior art, TM below 0.5) are the GPU compute we scoped and deferred. Anchored on, not copied from, the published UMG-SP2 urethanase transition-state geometry.
- Engine C, Digital Recyclability Passport. A live compliance classifier that routes each garment clear / lab-test / divert against real US and EU regulation (CPSIA, California PFAS law, REACH, OEKO-TEX, EU ESPR) and emits a scannable GS1 Digital Link / EU Digital Product Passport. Scanning the QR opens the live single-product passport on any phone, and each passport carries a real W3C Verifiable Credential you can cryptographically verify in the browser (Ed25519, with a tamper toggle, no server). The part Carter's could deploy in Year 1.
REWEAR-FUSED is one inverse-design loop, two coupled molecules. Rather than attack today's crystalline elastane (which defeats every enzyme), it redesigns the fiber to be cleavable and designs the enzyme to match.
flowchart LR
G["Worn onesie<br/>cotton + elastane"]
G --> C["Engine C<br/>Recyclability Passport<br/>clear / lab-test / divert"]
C --> A["Engine A<br/>cleavable elastane<br/>(constraint-guided screening)"]
A -- "carbamate substrate target" --> B["Engine B<br/>carbamate hydrolase<br/>(LigandMPNN redesign)"]
B -- "cleaves the bond" --> M["Recoverable monomers"]
M --> N["New onesie<br/>re-spun"]
N -.-> G
The product is wired so the live demo never depends on a real-time GPU job. The pipelines pre-compute signed artifacts; the deployed app renders them through real code; Engine C also runs live.
flowchart TB
subgraph CLIENT["Browser, the judge"]
FE["Next.js 16 app on Vercel<br/>5 views · Mol* · R3F · scannable QR"]
end
subgraph DATA["Signed artifact bundle · data/artifacts/v1.0.0"]
ART["enzymes · fibers · pairs · passports · pdb<br/>manifest.json with SHA-256"]
end
subgraph BACK["Pipelines + live endpoint"]
P["Engines A+B (Pravin)<br/>LigandMPNN to Boltz-2 to PLACER to FoldSeek (GPU; de-novo RFdiffusion deferred)"]
V["Engine C (Vinh)<br/>FastAPI on Render · live /classify"]
end
P -- "pre-computed, frozen" --> ART
V -- "serves + frozen-model parity" --> ART
FE -- "reads (demo-safe, no live GPU)" --> ART
FE -- "live classify" --> V
We would rather you check than take our word. The full tiering is the in-app /judges page.
| Component | Tier | TRL | What it is |
|---|---|---|---|
| Engine C, the Passport | WIRED LIVE | 6 | a real FastAPI endpoint + the deployed UI; the Year-1 deployable product |
| The application (5 views) | WIRED LIVE | - | deployed on Vercel, reading the signed artifact bundle through real code |
| Training + reference data | REAL DATA | - | 631 real CPSC recalls (saferproducts.gov), wired today; EU Safety Gate, Peaslee PFAS, OEKO-TEX are the roadmap |
| Engine A, the fiber | IN-SILICO | 3 | constraint-guided screening; predicted properties with benchmark-referenced confidence |
| Engine B, the enzyme | IN-SILICO | 2 | LigandMPNN redesign (TM ~0.8 to template); PLACER preorganization + FoldSeek novelty on the Lauko thresholds; the de-novo fold is the deferred swing |
| Wet-lab validation | FORWARD | - | not yet done in a lab; the Cox Cleantech Residency ask |
| View | What it shows | |
|---|---|---|
| 01 | Story | the two walls and the verified stats, as a scroll-driven cold open |
| 02 | Passport | garment intake, the clear/lab-test/divert router with citations, the scannable QR |
| 03 | Fiber (Engine A) | the screening funnel, the candidate, and the mechanics-vs-cleavability trade-off curve |
| 04 | Enzyme (Engine B) | the Mol* structure with the Ser-His-Asp triad, the PLACER metric, the FoldSeek badge |
| 05 | Loop | the GPU particle storm: onesie to cleavage to monomers to a new onesie, scrubbed by scroll |
Screens in docs/gallery/.
Three views of the project, broad to specific: a git-history time-lapse, an auto-generated codebase graph, and a hand-curated interactive 3D knowledge graph. Full index in docs/visualizations/.
- Git-history time-lapse (Gource):
docs/visualizations/rewear-fused-gource.mp4. Click the still above to play. - 3D knowledge graph (interactive, hover + click): rewear-fused.vercel.app/visualizations/knowledge-graph.html: the co-design loop and the two coupled molecules, glowing on a black canvas.
- Codebase graph (graphify, 287 nodes / 401 edges): live ·
GRAPH_REPORT.md.
# Frontend (Next.js 16, webpack bundler for Mol* compatibility)
cd frontend && npm install && npm run dev # http://localhost:3000
# Engine C, the live passport endpoint
cd engine-c && python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt && pytest -q
uvicorn api.main:app --port 8000 # GET /healthz, /classify/{id}, /passport/{id}
# Regenerate the artifact manifest (SHA-256 per file)
python scripts/build_manifest.pyEngines A and B run on a rented GPU (RunPod). Deploy is to Vercel (frontend) + Render (Engine C).
frontend/ Next.js 16 + React 19 + Mol* + R3F. The five views + the scroll shell + /judges.
backend/ Engines A + B molecular pipelines (Pravin) + the artifact server.
engine-c/ Engine C: the compliance classifier, the passport, the live FastAPI endpoint (Vinh).
data/ The versioned, SHA-256-signed artifact bundle the frontend reads. The data contract.
scripts/ build_manifest.py (re-sign the bundle), validate_artifacts.py (contract gate), sync_artifacts_to_frontend.py, download_pdb_placeholders.py.
docs/ The deploy guide and the screenshot gallery.
Next.js 16, React 19, TypeScript, Tailwind v4, Mol* 5.9, three.js r184, React Three Fiber 9, GSAP, Lenis, visx, FastAPI, Python 3.12, Render, Vercel, RDKit, LigandMPNN, Boltz-2, PLACER, FoldSeek, CatBoost, GS1 Digital Link, W3C Verifiable Credentials.
REWEAR-FUSED is decision support and a design tool, not a validated material or a medical claim. Everything is in-silico at TRL 2-3 and labeled as such on every predicted-property panel. Engine C trains on real public data only; no synthetic data reaches the judged path. Demo garments are real Carter's product archetypes with real published compositions, and no fictional persona is used. Aliphatic-only chemistry is enforced so no aromatic carcinogen can be released. Every claim is checkable in the code and on the /judges page.