S">S"> Sivakumar Portfolio
Sivakumar Portfolio
Computer Vision · Mobile · Patent Filed · Field Research

Nissan Connect
MIL Diagnostics

CV-based malfunction indicator light diagnosis. FPNLite on-device detection, AR overlay, VIN-based severity classification. The product that needed 20+ garage visits across three cities before a single line of spec was written.

FPNLite · On-device MLAR Core · Label Overlay VIN-based SeverityPatent 202441039454
Detection Model
FPNLite · on-device inference
Users
17,000 · 6 months post-launch
Field Research
20+ garages · Chennai · Bengaluru · Hyderabad
Outcomes
70%
unnecessary service visits
eliminated · post v2
4.5
CSAT out of 5.0
driver self-resolution
vs pre-launch
−64%
false positive rate
v1 → v2
91%
MIL symbol detection
FPNLite · on-device
Field Research Discovery

The team's diagnosis was wrong. Post-launch, unnecessary service visits were increasing. Internal consensus: user education problem to drivers don't understand MIL warnings. My diagnosis after 20+ garage visits across Chennai, Bengaluru, and Hyderabad: system over-signalling problem. The platform was flagging transient electrical faults (vibration, temperature variance, loose contacts) as persistent critical faults. Drivers were behaving rationally to responding to a broken signal. Fixing user education would have been expensive and wrong. The v2 pattern detection fix took one sprint.

Design Architecture
System design architecture  to  FPNLite detection pipeline to severity classification

System design architecture to FPNLite detection pipeline to severity classification

End-to-end customer journey  to  MIL trigger to resolution

End-to-end customer journey to MIL trigger to resolution

Model Training
FPNLite model training pipeline  to  12,000+ images · 6 lighting conditions

FPNLite model training pipeline to 12,000+ images · 6 lighting conditions

System Architecture
Vehicle Identification
QR code scan · VIN embeddedVIN decoder · Nissan parts manifest APIVariant lookup · base/mid/premium/sport75 to 85 checkpoints per variant
Computer Vision Detection · FPNLite
Camera feed · real-time frame captureFPNLite model · MIL symbol detection · on-deviceAR Core · digital label overlayConfidence threshold · 0.75 · below → photo fallback12,000+ training images · 6 lighting conditions
Pattern Detection · v2 Fix
3-occurrence threshold · 15-min window requiredVoltage variance filter · ±0.2V transient suppressionSeverity classifier · Critical / Safe-to-Drive / Self-ResolvableOBD-II telemetry · fault code validation
Resolution Layer
Critical → immediate dealer call + RSASafe-to-drive → monitor + self-resolution guideSelf-resolvable → step-by-step in-appNearest dealer · Google Maps API
App Screens
Landing  to  malfunction detected
QR scan → VIN lookup → diagnosis
Severity classification + recommended action
Nearest dealer  to  Critical severity path
Roadside assistance  to  emergency path
Diagnostic reports  to  post-resolution tracking
Patent

Patent filed: Intelligent Systems for Identifying Malfunctions to Application ID: 202441039454 to Status: Filed. Covers VIN-based variant lookup combined with on-device CV detection and severity classification logic.

Patent wall  to  Application ID 202441039454

Patent wall to Application ID 202441039454

The v2 Architecture Fix
v1 Logicv2 LogicImpact
Single fault → persistent flag3 occurrences in 15-min window requiredFalse positives −64%
All MIL signals treated equallySeverity classifier · Critical/Safe/Self-resolvableUnnecessary visits −70%
Text description onlyAR Core overlay · visual label on physical symbolSelf-resolution rate 2×
Voltage variance flagged as fault±0.2V transient filter appliedVibration false positives eliminated