How it works

SciRAG is a guided wizard, not a chat window. It asks for what a risk assessment actually requires, checks the chemistry before it writes anything, and produces a document in Safe Work Australia format with every claim traceable to a source.

It supports both chemical and biological assessments. Here is what happens at each stage.

You describe the experiment

There are two ways in. You can enter the details directly through a short form, or you can upload an existing method document, whether that is a paper, a thesis chapter, an electronic lab notebook entry, or a protocol. SciRAG reads it and extracts the experimental parameters for you to confirm.

Alongside this you upload the Safety Data Sheets for your reagents. The first time a given SDS is seen, it is parsed into a structured chemical record. After that it is drawn from the knowledge base rather than read again, which keeps generation fast and holds the cost down.

The experiment details form filled in for a Grignard synthesis: reaction name, solvent, temperature, atmosphere, equipment, and reactive reagent categories.

SciRAG runs a safety pre-check before it drafts anything

This step exists because the most dangerous thing about a risk assessment is the hazard nobody thought to look for.

Before generation is allowed to begin, SciRAG checks every reagent against every other for genuine reactive incompatibilities: combinations that produce a violent exotherm, an explosive peroxide, or an acutely toxic gas. It flags conditions that are hazardous in themselves, not merely chemicals that happen to share a hazard class.

Where an incompatibility is rated Critical or High, generation stays locked until you state how you intend to control it. Your answer is not a formality. It is passed into the assessment and reflected in the control measures. If a procedure cannot be conducted safely as designed, SciRAG is permitted to return a No-Go determination rather than write around the problem.

The safety pre-check panel showing a real HIGH-severity finding for a Grignard addition: a violent exotherm risk between methylmagnesium bromide and methyl butyrate, with recommended controls.

The assessment is generated in Safe Work Australia format

The output is a complete document, structured to the sections an OHS officer expects: hazard identification, a chemical table, incompatibilities, risk ratings, the hierarchy of controls, emergency procedures, and waste disposal. No reformatting is needed before submission.

Control measures and first aid are written in plain language a researcher can actually follow at the bench, rather than as regulatory boilerplate. Factual claims carry a citation marker back to the source they came from.

The hazard identification and risk rating section of a generated risk assessment, with a High-risk card each for methylmagnesium bromide and diethyl ether, and a Medium-risk card for methyl butyrate, each with GHS classifications and source SDS sections cited.

A verification pass audits the draft

Once the assessment is on screen, a second pass reviews it critically. It looks for claims that are not supported by the uploaded sources, internal contradictions, and exposure limits or hazard statements that appear to have been assumed rather than read from an SDS.

Anything the model drew from its own training data rather than from your documents is marked as such, so you know precisely which statements deserve a closer look. The aim is not to tell you the draft is fine. The aim is to tell you where it might not be.

The AI-generated-content disclaimer above the accuracy check result, which passed with no issues identified for this assessment.

You review, and your institution approves

SciRAG drafts. It does not sign off. Every assessment is reviewed by the researcher, then passed through the approval chain your institution already runs, typically the lab head and then the departmental OHS officer.

This is deliberate and it is not going to change. The platform is built to give a reviewer a better document to review, not to remove the reviewer.

What grounds the output

SciRAG works from what is in front of it. If a chemical or a scenario is not covered by the available sources, it will say so rather than invent an answer.

Safe Work Australia: The primary compliance framework. Output follows SWA format and applies the hierarchy of controls.
NHMRC guidelines: National Health and Medical Research Council biosafety guidance, used for biological assessments.
GHS and IARC classifications: Globally Harmonised System hazard codes, and IARC carcinogen designations in Groups 1, 2A and 2B.
Open-Source Chemical SDS Database: A curated database of Safety Data Sheets, parsed into structured records at the point of generation.

See it on your own experiment

Bring a method and a handful of Safety Data Sheets, and have a draft assessment in front of you in about ten minutes.

Create an account