Chemistry × AI

Draw a molecule. Let AI predict its spectra.

Enter a structure and get predicted ¹H NMR, ¹³C NMR and IR spectra — with charts, peak tables and atoms that light up when you click a peak.

Start predicting See how it works
3
spectrum types
4
ways to enter a molecule
2
AI providers
Schematic: a molecule goes through an AI model and becomes a predicted spectrum CCc1ccccc1 C₈H₁₀ · 106.17 AI AI model ¹³C NMR Predicted
Schematic illustration — ethylbenzene

Input

Start with whatever you have

Four ways to give the app a molecule — every one ends in the same checked, confirmed structure.

SMILES

Paste a SMILES string. It is parsed and canonicalised instantly.

CC(=O)Oc1ccccc1C(=O)O

Draw it

Sketch the structure in the built-in molecule editor.

MOL / SDF file

Upload a molfile or an SDF and the structure is read for you.

Structure image

Upload a picture; the AI reads it and you confirm the result.

How it works

From structure to spectra in four steps

A simple flow, and every step is under your control.

  1. 1

    Enter the molecule

    Type a SMILES, draw it, or upload a file or image.

  2. 2

    Confirm the structure

    RDKit validates it and shows the canonical SMILES, InChIKey, formula and mass. The prediction only starts after you confirm.

  3. 3

    Choose spectra and conditions

    Pick ¹H, ¹³C and IR, the NMR solvent and spectrometer frequency, the IR method and the AI model.

  4. 4

    Explore the result

    Charts, peak tables and atom highlighting for every spectrum.

Live demo

Click a peak, see the atoms

Every peak is linked to the carbons that produce it. Try the ethylbenzene ¹³C example below.

Try it with your own molecule

Spectra

Three spectra, one workflow

Choose any combination — each one is predicted separately.

¹H NMR

Chemical shift, multiplicity, integration and coupling constants on a 12 → 0 ppm axis.

¹³C NMR

One peak per unique carbon, with its assignment and the carbon environment.

IR

Wavenumber, intensity and the functional group behind each band — ATR, KBr, film or solution.

Solvent and spectrometer frequency are yours to set (CDCl₃, DMSO-d₆, D₂O … 300–600 MHz or custom).

Honest by design

Predictions you can read critically

Built for exploration and teaching. The app never dresses an estimate up as a measurement.

Always labelled as predicted

Every result carries a clear notice that the spectrum is an AI prediction, not laboratory data.

Predicted spectrum — not laboratory data.

No invented numbers

When the model gives no multiplicity, coupling constant or confidence, you see “Not predicted” — never a made-up default.

J (Hz)Not predicted

Structures are checked first

RDKit validates every molecule, and the AI is given the verified atom list, so peaks map to real atoms.

Your model, your keys

Choose OpenAI or Claude per analysis. API keys stay on the server and never reach the browser.

One failure does not stop the rest

Each spectrum runs on its own. If one fails the others still appear, and you can retry just that one.

¹H NMRCompleted
¹³C NMRCompleted
IRFailedRetry

Ready to see your molecule's spectra?

Open the predictor, enter a SMILES and watch the spectra appear.

Open the predictor