Molecular Photophysics AI
TD-DFT takes 24 to 72 hours per molecule, so millions of virtual candidates get narrowed to a few hundred that anyone actually evaluates. Our graph isomorphism network predicts absorption, emission, bandwidths, extinction, quantum yield, and lifetime straight from a SMILES string.
Predicted in PBS, DMSO, ACN/water, ethanol, and hexane. Solvent-conditioned GIN, because the same molecule behaves differently in two buffers.
The problem
Before anyone synthesises a candidate, someone needs to know how it absorbs, emits, and decays. Today that means TD-DFT. It is accurate, and slow enough that only a sliver of any real virtual library ever gets checked. The rest is guesswork, or it never gets screened at all.
Try the model
The preview below is scripted, using four reference fluorophores. The production model accepts any valid SMILES string. Try the model for live predictions.
How it fits your workflow
A two-stage pipeline: cheap, fast pre-screening across the full virtual library, then TD-DFT validation only on the candidates that clear the bar.
SMILES in, seven properties out, in under a second per molecule. Rank a virtual library that would take decades to touch with TD-DFT.
Spend your quantum-chemistry budget on the survivors. Complementary to the tools you already run, not a replacement for them.
At $5K to $20K and weeks of lab time per candidate, the cost of the pipeline is decided long before anyone touches a flask.
Test-set performance
Held-out R² across all seven predicted properties, with 697 to 1,850 test measurements per property. We publish all seven, not just the best ones.
From predictions to decisions
Every prediction is classified against thresholds you set for your own use case. Turn seven predicted properties into one flag your bench team can act on.
Four workflows, four buyers
Different molecules, same problem: optical properties gate every downstream decision, and today they're found out too late.
For QC and analytical chemistry heads at CDMOs and generics companies.
Plan UV detector wavelength and concentration range from one SMILES string at project intake, before synthesis. Cut the 2 to 4 week post-synthesis delay on every new API.
For assay development teams at CROs, biotechs, and vaccine manufacturers.
Pre-screen a whole dye catalogue in minutes. Match emission, quantum yield, and lifetime to your plate-reader filter set and assay format before any procurement.
For display materials R&D at emitter developers and panel makers.
Rank candidates by colour gamut, quantum yield, and lifetime in the microsecond TADF window, simultaneously. Screen 100 candidates in roughly 100 seconds, before committing to synthesis.
For conjugate development teams at pharma and biotechs.
Predict payload extinction coefficient and payload-specific max wavelength before HPAPI synthesis. The inputs every DAR calculation needs, ready before the first conjugation.
SMILES in, predictions out
We're pricing this with our earliest customers. Tell us your volume and we'll find a number together.
Blind Retrospective Validation: send us 20 molecules you've already measured, whether shelved, historical, or non-active. We predict all seven properties blind. You compare against your own records.
Direct to the founders. No procurement, no IT onboarding.
Who we are
UAE · India · Global Customers
Your molecules are your IP. Contact the founders to discuss data-handling terms for your specific engagement.
Graph Isomorphism Network (GIN), as powerful as the Weisfeiler-Lehman test for distinguishing molecular graph structure. Trained on 20,796 experimental chromophore-solvent measurement points rather than TD-DFT theoretical data.