PRISMby Mirdyne

Diagram: The five PRISM stacks, Research on top and Evidence as the foundation

01The platform

What PRISM is built from.

PRISM has five stacks, each with one job. Every part is marked with how ready it is.

  • In use Used in our projects today
  • Prototype Working, being improved
  • In development Being built
  • Target The goal

Stack 01 / 05Research

Stack 01 / 05

Research stack

Out of millions of possible mixes, which are worth making?

PRISM ranks a short list of candidates and says how sure it is about each. Physics simulations throw out what cannot work before any powder is weighed.

  1. 01
    Knowledge graphPapers, patents and lab data, linked, each with its source.
    Prototype
  2. 02
    GeneratorSuggests a whole batch of promising recipes, not one best guess.
    Prototype
  3. 03
    Judges and trust meterSeveral models score each recipe. Where they disagree, the score counts for less.
    Prototype
  4. 04
    Exact checkThe promising recipes the models are unsure about get an exact quantum calculation.
    Prototype
  5. 05
    A window, not a pointA safe range that still works when powder and machine change.
    Prototype

LimitIt cannot see what a real machine does to the alloy. The manufacturing stack checks that.

Stack 02 / 05

Harness stack

Who runs the work between the AI models?

The harness does. The AI models only make suggestions. The harness plans each round, runs the tools, scores the results and remembers what failed.

  1. 01
    PlannerDecides what to try next, from the last results.
    Prototype
  2. 02
    PlaybooksWhat each round learned, failures included.
    Prototype
  3. 03
    Tool connectionsOne way in to simulations, lab instruments and factory data.
    Prototype
  4. 04
    ScorerScores every result against the requirement.
    Prototype
  5. 05
    Human sign-offAI suggests, engineers decide. Ideas without a source are rejected.
    Prototype

LimitIt never makes the final call. A named engineer signs off what leaves the loop.

Stack 03 / 05

Autonomy stack

How do experiments stop being the slow part?

Robots do the repetitive steps, and people stay in charge. They weigh and heat, instruments measure on the spot, and the data goes straight back into the loop.

  1. 01
    Recipe writerTurns an idea into steps a lab can run.
    Prototype
  2. 02
    Robot labRobot arms weigh powder and move samples through the furnace.
    In development
  3. 03
    Automatic measurementX-ray patterns on the spot, read by AI.
    In development
  4. 04
    ProbesOur own sensors, recording calibrated data at the machine.
    In development
  5. 05
    Machine controlSoftware drives the instruments, so the loop never waits.
    In development
  6. 06
    Field robotsThe same autonomy outside the lab, without GPS. For civil and defence use.
    In development

LimitMost of this is still being built. The recipe writer works as a prototype today.

Stack 04 / 05

Manufacturing and test stack

Can it really be made? Does it hold up?

Only a real test can tell. Most materials designed on a computer never get past the recipe. We melt the best ideas, mill them down, make powder and print them, then test them.

A vacuum-arc furnace, open: the steel chamber with its viewports lifted above the round copper hearth.
The vacuum-arc furnace, open.
  1. 01
    Melting and powderMelted in a vacuum-arc furnace, milled down and made into powder.
    In use
  2. 02
    Printability checkCan it be 3D-printed? Checked before any build starts.
    Prototype
  3. 03
    Metal 3D printingLaser powder bed fusion, first at Fraunhofer IAPT, then on Bimo Tech’s industrial machines, with the safe settings carried across. DED too.
    In use
  4. 04
    TestingDensity and inner structure first, then strength and heat.
    In use
  5. 05
    Certification fileThe evidence certification needs, from test sample to real part.
    Target

LimitA test sample is not a finished part. Certification is our goal; we do not claim it yet.

Stack 05 / 05

Evidence stack

Where did this result come from, and who may see it?

Every result records where it came from, who owns it and who may use it. Requirements, designs, samples, tests and decisions are linked, like a family tree.

  1. 01
    TraceabilityEvery result records its inputs: data, code and model versions.
    In use
  2. 02
    Linked recordsRequirement to decision, linked instead of scattered in files.
    Prototype
  3. 03
    Data rightsEach piece of data carries its owner and its allowed uses.
    In development
  4. 04
    Controlled sharingNothing is shared by accident. Every release is signed.
    In development
  5. 05
    Export controlExport rules on every deliverable today; automatic checks next.
    In use

LimitTracing and export labels work today. Automatic enforcement is being built.

02Progress

What PRISM has done, and what comes next.

Running now · funded by ESA

PRISM Alpha

A project in ESA’s Future Launchers Preparatory Programme (FLPP), and the first built around the full PRISM loop, for European space transport.

The best candidates are melted, milled and made into powder, printed at Fraunhofer IAPT and then at Bimo Tech, and tested as coupons.

12 months
TRL 3 to 4
3+
candidate alloys
1
complete closed loop

Kicked off in July 2026. The award ceremony was held in October.

  • European Space AgencyCustomer and funder
  • Bimo TechPrime contractor
  • ArianeGroupRequirements and validation
  • Fraunhofer IAPTMetal 3D printing
  • amsightManufacturing data

Where it stands

Done so far

  • July 2025: we put PRISM to ESA through OSIP, its platform for new ideas: our submission (PDF).
  • Project SPARK: eight candidate alloys narrowed to two, made as real metal.
  • PRISM Alpha, a full ESA project in the launcher programme FLPP, kicked off in July 2026.
  • Our first privately funded project, for PFAS‑free polymers, is signed.
  • PRISM won the AI special prize (KI‑Sonderpreis) at Hessen Ideen 2026.
  • PRISM received a FIRST! Simulation & Intelligence award at the ESA FLPP Autumn Session in Paris on 14 October 2026.

Still to prove

  • One full loop, from requirement to test results. That is PRISM Alpha’s job.
  • Robots making samples and software driving the instruments, on a real line.
  • A material taken from test sample to real part, with the evidence certification needs.

Materials programmes

The loop, taken to new kinds of material.

  1. ActiveRefractory alloys
  2. ContractedPFAS‑free polymers
  3. NextPolymers and bio-based materials

Supply-chain intelligence

The same tools, used to see how a change at one supplier spreads.

  1. InternalSupply risk
  2. PrototypeMarket signals
  3. ExploratoryProgramme risk
  4. NextEarly warnings

How a disruption spreads

Pick the supplier where trouble starts, then watch it spread to the programmes that depend on it.

Diagram: Illustrative supply network from mines to programmes, with feedback from the programmes back to the mines.

Chance each programme is hit

  • Launcher24%
  • Jet engine20%
  • Fusion reactor16%
Impact peaks after
31 days
1 in 20 disruptions reach
49+ events

We use the same model to forecast how a material degrades: sensor readings become events, and the model predicts when a part will fail.

Illustrative

The network and its rates are made up; the chances are computed exactly from the model, a Hawkes process. Our methods are adapted from Okawa et al., “Dynamic Hawkes Processes for Discovering Time-evolving Communities’ States behind Diffusion Processes”, KDD 2021, for supply chains, and from Repasky et al., “Modeling Discrete Coating Degradation Events via Hawkes Processes”, 2025, for degradation.

Whatever PRISM is used for, every claim keeps its source and says how sure it is.

Start

Tell us what your part must survive.

We’ll tell you what PRISM can find, and how we’d prove it.

Register interestGitHub