Predict. Print. Prove.
Repeat.
Three integrated platforms operate as a single closed development loop. Each one alone is useful; together they produce process-controlled training data that compounds over time.
CEMFORGE — Formulation Intelligence
CEMFORGE is an AI mix-formulation platform built on top of the Modified Andreasen-Andersen particle-packing model (Funk & Dinger, 1994), trained on validated 3DCP specimen data anchored to the open Open3DCP schema. It accepts your available raw materials — binders, aggregates, supplementary cementitious materials, admixtures — and your target performance properties, and returns candidate mix designs for lab validation.
The platform runs in the browser. Outputs are structured for direct hand-off to a print run on M3-CRETE or any 3DCP printer with documented process parameters. Available at cemforge.ai.
M3-CRETE — The Research Printer
M3-CRETE is an open-source pallet-scale concrete 3D printer with a target build cost under $5,000. CERN-OHL-W-2.0 hardware. The full CAD, BOM, firmware, and build documentation are public; the project is documented in real time on Hackaday.io.
Why open hardware: 3D concrete printing has been dominated by proprietary equipment with opaque print parameters. Field-wide research progress depends on documented, inspectable process control. M3-CRETE is engineered for that role specifically — not as a turnkey on-site house printer, but as a research-and-development platform whose every variable is available for inspection.
See the printer at /m3-crete/ and the source repository on GitHub.
ACME Lab — Autonomous Validation
On our three-year roadmap is the ACME Lab — Autonomous Cementitious Materials Engineering Lab — a dedicated facility designed to run the predict-print-prove cycle at higher throughput with greater automation, from mix batching through print, cure, and destructive test.
The current lab-scale work demonstrates the value of process-controlled training data. ACME Lab is intended to scale that value — not create it from scratch. Fundraising is underway. If you are a research institution, agency, or investor interested in accelerating 3D concrete printing development infrastructure, contact us at [email protected].
The RILEM Evidence
The RILEM TC 304-ADC inter-laboratory study on 3D-printed concrete mechanics had 30 labs each print and test their own mix under a shared protocol. Per-lab 28-day printed strengths spanned roughly 27 to 96 MPa — a spread driven by materials and printing process together, and one the field cannot decompose because process parameters were rarely recorded. Printing process belongs in the dataset, not in the noise.
For AI training: if your training data mixes results from many labs with different unrecorded process parameters, you are training on process noise as if it were materials signal. Sunnyday operates the formulation engine, the printer, and the test program. Every specimen retains the process parameters that produced it. That coupling is the data asset.
Read the methodology paper
Convergence Engineering — the long-form whitepaper behind the closed-loop approach — is open-access on the research hub.