01 · BRIEF
Start from what you have
Paste a spec sheet, an email thread or last quarter's project notes. ALAi pulls out the requirements, asks about the gaps and shows where each one came from.
Molecule discovery for industrial chemistry
Bring us a constraint.
We'll bring you a molecule.
Drop-in molecules for the plants industry already runs. Designed by a pipeline built for your chemistry, proven in testing, delivered with the evidence attached.
CHEMIST
DR. ELENA VANCEMatCo R&D
Search molecules and briefs
EV
Good morning, Elena
1 cohort in flight · triage below
Needs you 3
Outcomes returned: Thermal IL cohort ATest results returned · calibration delta ready for review
OPENApprove handoff for Thermal IL cohort AAgent checkpoint · Autopilot drafted the work order. Review and send.
REVIEWDiscovery run finishedRun complete · 3 candidates vetoed by Safety guard. Review cohort ranking.
VIEWNew brief · step 1 of 3
Paste what you have
THE ENGINE WILL EXTRACT REQUIREMENTSExtract slots
viscosity ≤ 150 cP
flash point > 200 °C
Al6061 / EPDM compatible
not REACH SVHC
cost < $50/kg
Thermal-management IL for battery cooling loops
Thermal IL cohort A
3 candidates vetoed by Safety guardStructural alerts are never ranked. Safety is a hard veto, not a trade-off.
RankCandidate5-Lens
#1IL-AMN-004P 125 cPM spec'dS 1Su spec'dD 1Open
#2IL-AMN-002P 134 cPM spec'dS 1Su spec'dD 1Open
#3IL-AMN-006P 142 cPM spec'dS 1Su spec'dD 0Open
#4IL-AMN-005P 187 cPM spec'dS 1Su spec'dD 1Open
Cohorts · Thermal IL cohort A
Validated set
Outcomes returned. Validated molecules move to the Evidence Room; misses recalibrate the model.
2Validated
1Invalidated
1Pending
MoleculePredicted → measuredVerdict
IL-AMN-004125 cP → 118 cPIn windowEvidence
IL-AMN-002134 cP → pendingPending
IL-AMN-006142 cP → 171 cPOutRecalibrate
IL-AMN-005187 cP → 212 cPIn windowEvidence
Library · Evidence Room
5-Lens Certificate
Licensed molecule
IL-AMN-004
MatCo R&D · pipeline: thermal-IL v3 · GNN ensemble · retrosynthesis · process sim
Print PDFJSON sidecar
P
PerformanceViscosity, stability and heat transfer, calibrated on 84 returned outcomes
M
Manufacturability3-step route from commercial materials · ~$42/kg at tonne scale
S
SafetyGHS Cat 4 · no PBT flags · not on REACH SVHC
Su
Sustainability64% OECD 301B biodegradation at 28 days
D
Drop-inCompatible with existing Al6061 / EPDM loop · >200 h pump shear
Signed: mv_matco_adapter · sha256 reproducibility hash
Trust Center
Live view of who has access to your workspace, on what authority, for how long.
Active ALAi accessNo live deep-view sessions. Only aggregate, anonymized telemetry is visible to ALAi right now.
Require my approvalBreak-glass sessions need a live acknowledgement.
Freeze workspaceBlock all access, including support.
Did my data train a shared model?Cross-tier audit query, signed.
No · proof
Proven with
Independently reproduced by
URV
Independently reproduced by
URV
The problem
Industry runs on chemistry chosen decades ago.
Plants are built to run 40 years on the same molecules. A refinery costs billions and can't be rebuilt, only improved. So the chemistry inside it rarely changes.
10–20yrs
to take a new industrial molecule from idea to plant.
$100M
spent on a single molecule, at the high end.
90%
of candidates fail along the way.
1060
Possible molecules. Humanity has catalogued 220 million.The molecule your plant needs probably exists. Until now, nobody could search for it.
The platform
From constraint to certified molecule, in one workspace.
The ALAi workspace turns your operating limits into ranked, tested, documented candidates. Your team decides at every step.
02 · DISCOVER
Search the space, not the catalogue
Hundreds of thousands of candidates, scored on five lenses: performance, manufacturability, safety, sustainability and drop-in fit. Unsafe molecules are removed, not ranked lower.
03 · VALIDATE
Predicted vs. measured
Top candidates go to testing. Results land next to the prediction, misses included, and every result recalibrates the models for that chemistry.
04 · CERTIFY
A certificate, not a slide
Each molecule ships with a signed 5-Lens Certificate. Every number traces to a test result and exports as PDF and data.
Most AI chemistry
Generates molecules.
Generation is now the easy part. A list of a million candidates is a research project, not an answer.
ALAi
Delivers the few that work in your plant.
Built for the hard part: narrowing to the candidates that fit your equipment, cost and safety limits, then proving them.
How the engine works
The right model for each chemistry.
ALAi is not one general-purpose chemistry model. It is a modular pipeline: each stage uses the method best suited to the job, and better models replace older ones as they ship. Every handoff is recorded, so a chemist can see what was predicted, by which method, and how it held up.
01
Constraints
Your operating limits, equipment and regulations become the rules of the search.
Requirements extraction · plant data02
Generate
Candidates built for the chemical family in question, not drawn from a general catalogue.
Generative models, tuned per family03
Predict
Specialist property predictors score every candidate, with uncertainty attached to each number.
Graph neural network ensembles04
Make
Can it be made from commercial materials, and at what cost per kilogram?
Retrosynthesis planning · cost models05
Verify
Physics-based checks on the shortlist, before anything goes to a lab.
Quantum-chemistry cross-checks06
Scale
Does it work in the plant, not just the flask? Simulated at process scale, then tested.
Process simulation · lab validationModel-agnostic by design. When a better predictor ships, it replaces the old one at its stage. Your constraints, data and history carry over.
Product tour
See the ALAi workspace in 60 seconds.
Results · CO₂ capture with SOCAR
A drop-in solvent for refinery carbon capture.
Refineries capture CO₂ with amine solvents that corrode equipment and burn energy to regenerate. ALAi designed an ionic liquid that runs in the towers SOCAR already has. No new equipment.
405,891→1,470→8
Molecules generated·Ranked·Finalists
67%
of CO₂ captured at 15 bar on SOCAR's digital twin of the STAR refinery.
43%
less energy to regenerate than the incumbent amine.
$15M
a year in operating savings per refinery, in SOCAR's own model.
95%
prediction accuracy, independently reproduced by URV.
Our team was admittedly skeptical. After running our quantum modeling, we were genuinely astonished.Dr. Alberto Puga · URV, Universitat Rovira i Virgili
Capture performance simulated on SOCAR's Aspen digital twin. Savings from SOCAR's modeling. Accuracy reproduced by URV against quantum modeling. Read the case →
Applications
One engine. Every molecule a plant depends on.
CO₂ capture is where we proved it. The same engine takes any problem where the chemistry has to change and the plant can't.
Carbon capture
Solvents that capture more CO₂ on less energy, in the absorption towers already installed.
Separation and recovery
Pull contaminants and value out of produced water and process streams.
Thermal fluids
Coolants for battery packs and power electronics that move more heat at lower toxicity, in the loops and pumps already running.
Regulated chemistry replacement
Replace additives under regulatory pressure with safer molecules that pass the same performance tests.
Trust
Your data stays yours.
In chemistry, the molecule is the business. ALAi is built so your IP stays with you.
Your own pipeline
Your data fine-tunes your own copies of the models. It never trains a shared one.
You hold the keys
Support access needs your live approval. Freeze the workspace at any time, including to us.
Proof on request
Ask whether your data trained a shared model and get a signed answer, not a promise.
Start a brief
Industry won't rebuild itself to save the planet. We stopped asking it to.
Tell us what has to change and what can't: a performance target, a regulation, a cost ceiling, the equipment that stays. We'll come back with what's possible.