Two chemists, one rule.
Measure, or don’t ship.
A modest, disclosed correlation. Not a miracle.
Calibrated on known actives, and the interval is always shown.
We say so when a molecule is unlike anything the model has seen.
Anyone can claim rigour. We publish the places we are weak, in the same table as the places we are strong.
Four of these five made our own numbers worse. We shipped them anyway.
Near-constant, and correlated with lifespan at about zero. Dropped.
Our own standardizer was handing the test a free perfect match.
It rated aspirin 0.958 against BCL-2. Rebuilt.
Brand-new molecules were being labelled "already known". Fixed.
3D shape and electrostatics, beaten by plain 2D similarity.
Nobody made us do any of this. Every one was found by us and could have been left alone. A number that only ever moves upward is not a measured number.
From a target to something a chemist can make.
You pick a target and a tissue
Twenty-one targets across two programmes. Ten tissues, or whole-body.
→We build molecules, not pictures
Each one assembled from real building blocks through named reactions, so a route exists by construction.
→We score them on aging biology
Twelve hallmarks, compounds with measured lifespan effects, safety models and binding plausibility.
→You hand it to a chemist
Structure, route, ingredients and a stated confidence. A starting point, not an oracle.
We cite the published methods we build on. Our own calibrations stay ours, and can be reviewed under an NDA.
Small on purpose. You talk to the people who build it.
Builds the platform — the design engine, the scoring, and the product you sign in to.
Leads the chemistry — medicinal-chemistry judgement and what makes a candidate worth keeping.
Geroscience advisors — introductions welcome.
What we build on, and what stays yours.
Built on data anyone can check
DrugAge, ChEMBL, GTEx and the published hallmarks of aging. You can trace where every input came from.
Nothing for you to set up
No install, no procurement, no training run to budget for. Sign in and start.
The same run twice gives the same answer
Deterministic by design, with a registration hash per candidate and a versioned score.
Your chemistry stays yours
We never train on your inputs. Cloud, private VPC or on-premise; a data agreement is available.
