Chemical context before AI interpretation
ChemRotom is a macOS menu-bar prototype for chemistry and materials researchers. It adds a checkable step between a molecular structure image and an AI workflow: identify a candidate structure, inspect its chemistry, and compare it with the original image before copying the result.
The workflow
- Enter a candidate SMILES string, import a structure image, or capture a region of the screen.
- Use local DECIMER recognition to propose a structure when starting from an image; RDKit checks the candidate and generates chemical descriptors.
- Review an atom-numbered 2D preview alongside stereochemistry, unspecified configurations, and other warnings.
- Confirm the result, then copy the chemical context and structure image separately for use with AI.
Built for inspection
The SwiftUI interface connects to a local Python engine. Outputs include stereochemistry-preserving canonical SMILES, SELFIES, InChI, molecular formula, exact molecular weight, TPSA, Crippen LogP, and structured XML context. Image recognition and chemistry processing run locally; the prototype does not call an external vision or language-model API.
Current stage
This is a local development prototype for macOS 14+, with source available under the MIT license. There is no formal installer or signed release yet. Recognition dependencies and model weights require separate setup.
Chemical validity does not prove recognition accuracy. A structure accepted by RDKit may still differ from the original image. The current scope is clear, single-molecule 2D images; human verification remains essential. Broader literature-image evaluation and distribution packaging remain future work.