AI in Modern Safety Regulators
Jessie Nghiem, PhD
Energy Safe Victoria
Who are we
Energy Safe Victoria: the state’s energy safety regulator,
responsible for electricity, gas and pipelines safety.
Data and Analytics team:
- Team of 8
- Skillsets: data analysis, data modelling, BAU reporting and
advanced analytics
- In collaboration with Melbourne uni, Monash uni, Deakin uni
and CSIRO.
Featured on-going projects and future projects
1. Compliance check on e-commerce platforms
2. OCR enabled compliance check
3. Weather incident predictive model
4. Incident categorisation
5. Electrical Work on Online Service Platforms
6. Incident Forecasting model
7. Voice to text and text analytics and classifying the calls for triage.
8. RPA applications for web scraping/data pulling from online sales platforms
(eBay, Amazon, etc)
9. RPA for form processing
Compliance check on e-commerce platforms
Background
• 26% are now more willing to buy appliances online than prior to the
pandemic [Bain survey]
• 69% of total market revenue in Household Appliances segment will be
generated through online sales by 2024 [recent Statista report].
• The results of the latest stage of the project (involving 17 high-risk
categories) found 2,578 (30%) electrical products to be compliant from
8,555 listings that had a match in the database.
• The research component has been presented at ACM WSDM in Feb
this year (CORE A* conference in data science field)
• Each listing is matched to
EESS using the Model and
Brand (Trade Name)
• Model inputs:
Textual data for pattern
matching
Image data to determine if
the listing belongs to the
relevant category
How Does the Model Check for Compliance?
Online marketplace
How do we add value
The model facilitates audits by:
Reducing manual checks
Identifying irrelevant
listings (~70% not in-
category + excluded
listings)
Listings from Amazon, Catch and eBay
Total listings: 80,087
How do we add value
• A compliance officer could take ~6 years to check 100,000 listings
(approx. 50 listings / day). That can be done in a couple of hours with
this application.
Journey so far
Stage 5 –
Continuous Improvement
Discovery & Prototyping
Evaluate
Build & Iterate
Learn
Minimum
viable product
Audit products sold online
Audit products sold in-store
Crowdsource compliance
information for both online and
in-store audits
Stages 1 & 2
• Single
application to
audit online
sales
• 1 category pilot
Stages 3 & 4
• Identifying
opportunities and
challenges
• 31 categories
OCR enabled compliance check
Upload / click image
with brand and model information on the
web app
Figure 8 – OCR App Working
PC view Mobile view
Results
Featured list of projects and what’s next
Projects Status Complexity Funding
OCR enabled compliance check Ready to be
operationalised
Medium External
Compliance check on
eCommerce platform
Ready to be
operationalised
High External
Network-related incident
predictive model
Ready to be
operationalised
Medium Internal
Incident categorisation Ready to be
operationalised
Medium Internal
Electrical Work on Online
Service Platforms
Under development High Internal
Incident Forecasting model Under development High Mixed
Voice to text and text analytics Under development High Internal
RPA for data collection Not started Medium Internal
RPA for form processing Not started Medium Internal