Case Study
StarTex Software×Nanonets

From 10 minutes to 10 seconds
per Safety Data Sheet

StarTex Software's SDS library system required 10 minutes of manual extraction per Safety Data Sheet — reading 18 GHS sections, locating hazard codes, CAS numbers, and emergency procedures by hand. A Nanonets agentic pipeline now ingests, parses, and extracts all fields in 10 seconds with 96% confidence, auto-approving 90% of fields with no human touchpoint.

10min → 10sec/SDS90% Auto-Approved96% ConfidenceGHS Compliant
About the customerStarTex Software

StarTex Software builds EHS software for companies managing chemical safety and compliance. Their SDS library module needed to ingest Safety Data Sheets from hundreds of manufacturers — each in a different layout — and extract structured hazard data for the compliance database. Manual processing created a bottleneck that limited the system's scalability.

EHS Software
Product type
18
GHS sections/SDS
10 sec
Processing time
96%
Confidence score

The Problem

StarTex Software's SDS library module received Safety Data Sheets from hundreds of chemical manufacturers. Every SDS follows the 18-section GHS standard — but each manufacturer formats it differently. Extracting hazard statements, CAS numbers, and emergency procedures by hand took 10 minutes per document. With thousands of SDS documents to process, this created a compliance bottleneck.

  • 10 minutes of manual extraction per SDS — reading 18 GHS sections to locate key fields.
  • Varied manufacturer formatting — different fonts, layouts, and languages for the same GHS structure.
  • No structured output— extracted data couldn't be reliably loaded into the compliance database.
StarTex EHS
Upload SDS
Manual
Open document
Manual
Read 18 sections
Manual
Extract fields
Manual
Validate
Database
Manual entry
10 minutes manual per SDS document
18
GHS sections per SDS
10 min
Manual processing time
Stack
StarTex EHS App
GHS Standard
Excel

What used to take our team 10 minutes per SDS now takes 10 seconds. Nanonets reads all 18 sections, extracts everything we need, and flags the few fields it's unsure about.

Product Manager
StarTex Software
StarTex Software

The solution

One SDS agent. 18 sections. 10 seconds.

Nanonets replaced the manual extraction loop with one agentic pipeline. Four agents each own one step: receiving SDS uploads from the StarTex EHS application, parsing all 18 GHS sections regardless of manufacturer format, extracting hazard statements, CAS numbers, and emergency procedures with confidence scoring, and writing approved records to the SDS library — with flagged fields routed to human review.

Upload SDS File
Parse Document
Extract Fields
Quality Check
Update Database

Pipeline overview

StarTex App
SDS file upload
MSDS PDF received
Nanonets
Nanonets ingest
Format detected, queued
Nanonets
Nanonets extract
Hazard · Manufacturer · Specs
Nanonets
Quality flag
Confidence scored per field
Compliance DB
Compliance database
90% auto-approved
Library
Library updated
10 sec per document
Nanonets
Human review
10% flagged for verification
Loops back
After verification
10 min → 10 sec · 90% touchless · Millions of SDS documents processed

Impact

60×
faster per SDS
10 min → 10 seconds
90%
auto-approved
Fields auto-validated
96%
confidence
Extraction accuracy
9/10
satisfaction
Rated by StarTex
Metric
Before
After
Improvement
Processing time per SDS
10 minutes
10 seconds
60× faster
Auto-approved fields
0% (all manual)
90%
Touchless
Extraction confidence
N/A (manual)
96%
AI-validated
Section coverage
Manual scan
All 18 GHS sections
Full GHS coverage
Human review load
100% of documents
10% of fields only
90% reduction

AGENT ANATOMY

Nanonets Agent
Autonomous action-taking
After extraction
Agent decides path and acts
High confidence
All fields extracted
StarTex EHS
SDS uploaded
GHS format detected
Pipeline triggered ✓
Parser
18 sections split
Sections18 / 18
FormatGHS valid
Extractor
All fields extracted
Hazard: H302, H314 ✓
CAS: 1310-73-2 ✓
Mfr: BASF SE ✓
Quality Check
Score: ≥90% ✓
Confidence96%
StatusPass ✓
SDS Library
Record saved ✓
CAS1310-73-2
StatusSaved ✓
Flagged fields
Below threshold
Extractor
Field below threshold
⚠ Confidence: 62%
Review Queue
Routed to human
QueueReview
Nanonets
listening…
Reviewer
Field confirmed
StatusPending ⏳
SDS Library
Record saved ✓
Loops to saved
Parse failure
Unreadable section
Parser
Unreadable section
Section 3 unreadable ✗
Exception
Flagged for manual
StatusEscalated
Nanonets
listening…
Manual Entry
Fields entered
Entered manually
SDS Library
Record saved ✓
Loops to saved

See it run on your process, with your documents.

Start free. No credit card. Or talk to our team about your workflow.