Safety Data Sheets as computable, verifiable data.
OCSIP is an open, FHIR-inspired protocol and API that turns chemical Safety Data Sheets (SDSs) from PDFs into structured, rule-checked data. Every value links back to its source.
Get in touchThe problem
Every hazardous chemical shipped to a U.S. employer comes with a Safety Data Sheet. Structured formats exist, but most SDSs still arrive as PDFs that each receiver re-keys or re-abstracts, for every product and every revision.
AI can now read those PDFs, but it makes plausible mistakes. A value attached to the wrong ingredient, revision, or jurisdiction passes format checks and reaches the workers who rely on it.
The timing matters: under OSHA's 2024 Hazard Communication update, revised SDSs are due through 2027, so receivers are re-ingesting updated documents now.
Our approach
Like HL7 FHIR in healthcare, OCSIP exchanges linked, validated resources (substances, mixtures, hazard classifications, source documents, jurisdiction profiles) instead of whole documents. Underneath sits an open, executable model of GHS classification logic.
Rule-checked extraction
AI extracts each field. Every record is then checked against regulatory logic: label elements must follow from the stated classes, and classifications must be compatible with the disclosed composition. Reviewers see only what's suspect.
Evidence and provenance
Each value carries its source location, revision, and evidence status. Concentration ranges, withheld trade-secret data, and "not classified" statements are recorded as evidence states, not gaps.
Translation with loss accounting
When data moves between jurisdictions, each element is marked preserved, re-derived from evidence, or undeterminable, with the reason. Nothing is silently dropped or invented.
Who it's for
EHS and chemical-inventory software vendors
They ingest SDSs from thousands of suppliers and pay for manual abstraction.
Chemical distributors
They receive, store, and pass on SDSs for every product they carry.
Downstream manufacturers
They need current, traceable hazard data for worker safety and compliance.
The specification, reference implementation, validator, and conformance tests will be open. The commercial offering is a conversion API and review toolkit.
Status
OCSIP is in early development.
- Draft data model and executable rule model for U.S. (OSHA HCS 2024) and EU (CLP) hazard communication
- Benchmark on expert-annotated SDSs, comparing rule-checked extraction with extraction alone
- Open draft specification, REST/JSON reference server, and adapters for existing formats such as eSDScom XML
We're looking for design partners: EHS software vendors, distributors, and manufacturers who process SDSs at volume and want to share their requirements.
Contact
Interested in becoming a design partner, contributing to the specification, or learning more?