About SpectraScope
An observatory for atomic and molecular light — built to make spectroscopy approachable, fast, and a little bit beautiful.
What it does
SpectraScope handles six families of spectra: atomic emission/absorption (visible), infrared (IR), UV-visible, ¹H NMR, Raman, and mass spectrometry. You can upload your own data for AI-assisted identification, or browse the library of 15+ curated reference spectra.
How identification works
When you upload a spectrum, SpectraScope detects local maxima using a window-based prominence filter. The top peaks (or the ones you pin) are sent to an AI spectroscopist along with the spectrum type. The AI returns ranked candidate species and per-peak assignments in plain English.
The model is reminded of the spectrum-specific conventions: cm⁻¹ for IR/Raman, ppm vs TMS for NMR, m/z for mass spec, nm for atomic and UV-Vis. It's instructed to be explicit about uncertainty.
The library
The library ships with hand-curated reference data: hydrogen, helium, sodium, neon, mercury, iron; water, CO₂, methane, ethanol, benzene, acetone, caffeine, aspirin, and glucose. Only the spectrum types that are physically meaningful are shown for each species (e.g. CO₂ has IR and Raman, not NMR).
Data format
Two numeric columns: x (wavelength / wavenumber / ppm / m/z) and intensity. CSV with or without headers, or a JSON array of [x, y] pairs or objects like { "wavenumber": 1715, "transmittance": 0.18 }. Common alias columns (wavelength, wavenumber, ppm, mz, shift, absorbance, transmittance, abundance) are recognized.
Caveats
- This tool is for learning and exploration — not for clinical or forensic decisions.
- Peak detection is heuristic; very noisy or sloped baselines may need pre-processing.
- The AI can be confidently wrong on ambiguous spectra. Treat its output as a starting hypothesis.