Realise logo — European cat markRealiseEarly access →
[ Realise // v0.1 — Q3 2026 pilot ]

Your prototype works.
Your scale-up shouldn't fail.

An AI co-pilot for deeptech energy founders.

A physics-informed industrialization guide — from lab scale to production — powered by hybrid AI.

Feed us your engineering specs. We tell you what will break your scale-up — before you build it.

Physics-informedNot LLM-basedExplainable outputs
Scan your scale-up risk →See a live risk report →
€400K+

avg. rework cost avoided per project

18 months

avg. time lost given back

80%

early-detection target rate

[ Active failure modes ]

09 signals
  • Supply chain blindspot
  • 12–24 months lost
  • Hundreds of K€ burned
  • Lab scale → nowhere
  • Certification too late
  • Material delay
  • Design rework
  • Prototype works
  • Scale-up fails
[ The problem ]

Scale-up kills what the lab built.

01
· Trial-and-error costs a fortune

Design of experiments, protocol iteration, and data analysis cycles eat months and hundreds of thousands of euros — with no guarantee the next iteration works.

02
· Design lock-in comes too late

The window to freeze your design is narrow. Miss it and every downstream decision — supplier, tooling, certification — has to restart.

03
· Suppliers are an afterthought

By the time sourcing starts, lead times are already a problem. Negotiating quality control requirements mid-scale-up is a tax on your timeline.

04
· Regulatory dossiers arrive last

Compliance documentation is treated as a post-design task. It isn't. A certification requirement discovered after design freeze means a redesign — and months lost.

05
· The energy sector learns slowly

Learning rates in deeptech energy are long by nature. In a fast-moving market, not every solution can afford full trial-and-error cycles. Some need focused R&D — and a model that bridges the gap between what the physics says and what industry will accept.

[ The solution ]

Predict before you build.

12–24mo

typical time lost in the lab-to-scale valley of death.
Realise gives it back.

[ How it maps ]

From your design today
to scale-up scenarios.

One known state in. A trained inference graph in the middle. A spectrum of scale-up scenarios out.

—
—

Your design today

What you have today

  • MaterialsNi · Ir · PFSA
  • ProcessBench, 5–10 units
  • GeometryLab cell
  • TargetsTRL 4 perf.

Inference graph

Hybrid AI co-pilot

EncoderPhysics-informedConstraintDecoder
d₀=12d₁=64d₂=64dₙ=8Layer 1 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 1 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 1 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 2 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 2 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 2 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 3 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 3 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 3 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 4 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 4 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 4 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 5 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 5 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 5 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 6 · unit 1 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 6 · unit 2 — activation σ(Wx + b) routes the encoded condition through the inference graph.Layer 6 · unit 3 — activation σ(Wx + b) routes the encoded condition through the inference graph.ℒ
activationphysics prior weight |w|∇ℒ → θ
  • Supply chain graph
  • Cost & yield models
  • Manufacturing physics
  • Lead-time inference

Scale-up scenarios

Scenarios you can act on

  • Supply path

    EU-only · 6mo lead

  • Yield

    82% · ±3%

  • CapEx

    €1.4M est.

01 · Lab-to-Pilot

From bench to first production batch.

At 5–50 units the bottleneck is iridium availability and MEA assembly tolerance. Realise maps your bench process to validated EU suppliers and flags a 6-month lead time on Ir before you commit.

Time saved

8 mo

Safety & privacy

Realise surfaces scale-up scenarios, timelines, and risk flags based on your project inputs. It does not train on your proprietary data, expose underlying model weights, or share your information with third parties. All inferences are scoped to your session and discarded after the Condition Set is generated.

Analytics consent

Allow anonymous usage analytics (scenario views, interaction counts) to help us improve the copilot. No inputs, outputs, or identifying data are collected.

Status: Loading… ·

[ Who it's for ]

Built for the people building the energy transition.

Hydrogen electrolyzers

AEM · PEM · SOEC

Battery systems

Li-ion · Solid-state · Flow

Electrochemical components

Fuel cells · MEA · Bipolar plates

[ How deep do you need to go? ]

Three ways to scale with Realise.

Pro

Supply chain intelligence

EU supplier graph & lead-time mapping

Critical material availability (Ir, Ni, PFSA)

Sourcing risk alerts before design freeze

Best for: lab-stage teams preparing first pilot batch (5–50 units)

Request Pro access →
Premium

Supply chain + Manufacturing

Everything in Pro, plus:

Tolerance transfer analysis (lab → industrial scale)

Coating, assembly & yield prediction

Process window validation

Best for: pilot-stage teams scaling to demonstration (100–500 units)

Request Premium access →
Power

Full industrialization stack

Everything in Premium, plus:

Regulatory pathway mapping (ISO 22734, ATEX, DEKRA)

IP & freedom-to-operate landscape scan

EU funding window alignment — IPCEI, FCH2, HE — plus budgeting and business development IP strategy

Best for: demonstration-stage companies moving to industrial production (1,000+ units)

Talk about Power tier →
[ The founder ]

“Built by someone who lived every problem on this page.”

“From nuclear materials in India to AEM stacks in Grenoble — every role handed me the same lesson: the science works. The system around it breaks every time.
I built Realise because I was tired of watching it break.”

— Ronit Kumar Panda · Founder & CEO

[ The background ]

PhD · Electrochemical Engineering
Université Grenoble Alpes · 2024

Specialisations
materials engineering · polymer science · electrolyzer systems (AEM · PEM · SOEC) · degradation modelling · multiphysics simulation

[ The institutions ]

  • ·IGCAR Kalpakkam — nuclear materials research, India
  • ·IRSN — nuclear safety & materials, France
  • ·CEA Liten — national energy R&D institute, France
  • ·SOEC joint venture — solid oxide electrolyzer development
  • ·AEM consortium — IPCEI European hydrogen program
  • ·CUBE — deeptech incubation
  • ·31 European partners — coordinated across 8 countries

[ Expertise map ]

CentreRonit Kumar Panda

↳ Materials Engineering

polymers · electrodes · membranes

↳ Electrolyzer Systems

AEM · PEM · SOEC · stack design

↳ Scale-up & Industrialisation

TRL transitions · process · yield

↳ Supply Chain Intelligence

EU supplier networks · lead-time mapping

↳ Regulatory & Certification

ISO 22734 · ATEX · DEKRA · IPCEI

↳ R&D Coordination

31 partners · 8 countries · DemonHyc

“When Realise flags a supply chain risk or maps your certification pathway — it's not a generic model speaking. It's 10 years of those specific problems, encoded.”

See how it works →

[ Early access ]

Your next prototype
deserves better.

Join the waitlist. 3 pilot slots. Q3 2026.

Required fields marked with an asterisk.

0/1000

Supported by

BNP Paribas logoADIE logo