AI Connect
AI Compliance: Use Case from the Automotive Domain
22 August 2026 · RTU, Riga, Latvia
Assoc. Prof. Rosen Dimov, PhD
emotion3D / indie perspective · ACCOMPLISH showcase
Illustrations by Vladyslav Kulabukhov, @vladyskool
01
THESIS
Compliance as product evidence
02
Safety
Data
Software
Certification
Hazards
ODD · fallback
Lawful data
minimise · retain
Updates
secure · monitored
Audit trail
traceable · testable
Accuracy is not enough.
The claim must be evidenced.
EMOTION3D / INDIE
In-cabin monitoring
03
Driver state
gaze · drowsiness · distraction
Occupant state
presence · posture · child safety
Human context
hands · objects · engagement
AUTOMOTIVE SENSING
From cabin to situation awareness
04
Fusion
logic
Cabin
Telemetry
Radar / lidar
HMI
ODD · sensor limits · versioning · field feedback
ACCOMPLISH
Regulation to controls
05
Rules
AI Act · GDPR · Data Act
ASPICE · EURONCAP · etc
Operations
collect · prepare · apply · maintain
Indicators
measurable evidence
Controls
testable · auditable
Scorecards
review · certify
Human-centric assessment
living library of evidence
ACCOMPLISH
Reference architecture
06
Local toolkit
Platform layer
sensitive data
assessment engine
evidence collectors
governance
recommendations
digital passport
244 requirements · 30 components
SHOWCASE
Compliance-by-design workflow
07
1
Use case
2
Risk class
3
Evidence
4
Scorecard
5
Passport
6
Monitor
DMS / DOMS / shuttle monitoring
from model output to accountable release
AUTOMOTIVE EVIDENCE
What must be ready before deployment?
08
Data
representative · documented
Model
tested · versioned
Safety
ODD · fallback
Privacy
DPIA · minimised
Cyber
secure update
Market
drift · incident log
One chain: obligation → control → evidence → decision
THE LOOM: NOT YOUR SCRIPT
Chain of truth
provenance · verification · judgement
The AI system must not silently write reality.
“I write my own story.”
09