Chemical Verification Engine

A verification layer for AI-generated chemistry

Determines whether a generated procedure is internally coherent before experimental use.

Open reference implementation in development

Gitthub Link

AI Can Propose Chemistry — But Not Guarantee It

Modern language models and reaction predictors can create believable procedures.

But believable isn't the same as correct.

Their outputs may:

  • break reaction rules

  • leave out essential conditions

  • list incompatible reagents

  • look chemically reasonable yet be impossible

As automation grows, these mistakes become real experimental risks.

The Missing Step in the Workflow

Current:

Prediction → Human Guessing → Experiment

Proposed:

Prediction → Verification → Experiment

Chemical verification evaluates internal consistency and completeness before execution.

The engine extracts quantitative parameters and citations from a procedure and evaluates their cross-consistency. Instead of predicting outcomes, it evaluates whether the description itself is chemically coherent.

Verifying AI-Generated Chemistry

A deterministic evaluation of whether a generated procedure is internally consistent.

If prediction becomes automated, verification must become formalized.

Enables trustworthy autonomous experimentation and standardized evaluation of model outputs.