SDS Sentinel
Watch a folder, get audit-ready. No spreadsheets, no cloud uploads.
How SDS Sentinel Escaped The Idea Graveyard
A local-first Windows desktop app for EHS managers. It watches a folder for SDS PDFs, extracts GHS hazards, expiry, and SVHC, and gives you one dashboard and a one-click PDF audit report. Everything stays on your machine—SQLite, no cloud. Built for the gap between "Excel + shared drives" and big enterprise SDS platforms.
Genesis Spark
EHS teams drowning in SDS PDFs across folders and inboxes, and manual tracking that falls apart before audit season. I wanted something local-first: watch a folder, extract what inspectors actually care about (GHS, expiry, SVHC), and get one view and one report—no spreadsheets, no sending chemical data to the cloud. Solo micro-SaaS, built to live within a 5–10 h/week maintenance ceiling and a 1k MRR target (four customers at $249/mo).
Core Problem & Insight
The real problem isn't storing SDS—it's fractured tracking and last-minute audit panic. The bet: ingestion and verification can be production-grade (watchdog, parser router, consensus, audit trail) even when the product started as "four fields and a dashboard." Closing the gap meant adding GHS extraction (H/P-codes, signal word, pictograms), storage incompatibility from H-codes, SVHC cross-reference, a compliance score, and a real OSHA-style PDF report—so the app delivers audit-ready output, not just an indexed filing cabinet.
Known Chaos
- Parsing wildly different SDS PDF layouts and pulling GHS Section 2 (H/P-codes, signal word, pictograms) reliably
- Dealing with scanned SDS with no text layer (OCR)
- Keeping scope and maintenance low (solo, 5–10 h/week) while still adding hazard intelligence and a real audit report
Solutions
- Format-specific parsers (e.g. SpatialGHSParser with PyMuPDF) plus a router (parser/cache/OCR); static GHS reference for storage incompatibility
- Docling-based OCR behind a setting; router falls back when text parsing fails; clear "Parsed via OCR" and error feedback
- Strict scope: static SVHC list, no remediation workflows; GHS + SVHC + compliance score + PDF report as the launch set
Arsenal
Python
An interpreted, high-level and general-purpose programming language.
FastAPI
A modern, fast web framework for building APIs with Python based on standard type hints.
Jinja2
A fast, expressive, extensible templating engine for Python, often used for server-rendered HTML.
HTMX
A lightweight library that lets you build dynamic interfaces with simple HTML attributes and server-rendered partials—no heavy JS framework.
SQLite
A lightweight, file-based relational database engine that requires no separate server—ideal for local and embedded apps.
PyMuPDF
A high-performance Python library for reading, rendering, and manipulating PDF documents (fitz).
Docling
A document understanding library that converts PDFs and images into structured content, with OCR and layout analysis.
Watchdog
A Python library and shell utilities to monitor file system events in real-time.
pystray
A Python library for creating system tray icons and menus on Windows, macOS, and Linux.
fpdf2
A pure-Python library for generating PDF documents with minimal dependencies.
LemonSqueezy
A payment and licensing platform for digital products and SaaS, with APIs for subscriptions and license keys.
PyInstaller
A tool that bundles Python applications into standalone executables for distribution without a Python install.
Inno Setup
A free Windows installer creator used to build setup executables with a wizard-style UI and optional silent install.
Thanks for Peeking Behind the Curtain!
This one's still brewing in the lab. Head back and explore more chaotic creations!