Fsoft / SILABS IoT Challenge 2025 · 1st Runner-up · Leader of Edgelectronix

SCENT — Smart Customer Experience & Edge AIoT

End-to-end 24/7 AIoT system: MCU firmware (EFR32), Edge Gateway (Raspberry Pi 4), Flask backend, Redis + PostgreSQL, and NLP/Edge-CV pipelines for retail scent recommendations & inventory/threat monitoring.

Technical details Product Sketch Copy GitHub

Project Overview

SCENT is an offline-capable AIoT system for retail: personalized scent suggestions, inventory sensing, and environmental threat detection. Designed for continuous operation with local buffering & sync, and on-device/edge inference for low-latency experience.

Role

Team leader — Edgelectronix

Award

1st Runner-up · Internship Certification from FPT

Product Scope

Hardware → Firmware → Network → Software

Certifications

System Architecture

EFR32 Nodes
HX711, IR, I2S mic, SI7021

→

Raspberry Pi 4 (OTBR + MQTT)
Parsers, Smart LCD, MQTT broker

→

Desktop ( Redis / Flask + Postgres)
Workers, ETL, Software

→

AI & Data Pipeline
Datalake, Personalization & retrain

Product Stacks

EFR32xg24
AI perfume suggestion

EFR32xg26
AI drop detect

EFR32xg26
HX711 + IR

Raspi 4 + MG21
Open Thread Border Router

Desktop Admin
Software + DB

Firmware (EFR32)

System Flow

System Flow & Connectivity

Sensor Layer → Edge MCU:
- IR + Loadcell (HX711) → EFR32xG26 (Sensors) via GPIO interrupts
- SI7021 (Temp/Humidity) + Mic I²S → EFR32xG26 (Edge Drop Detection AI) via GPIO/I²S
Edge AI Processing:
- EFR32xG26 (sound/drop AI) → UDP results
- EFR32xG24 (Perfume Suggestion DL) → LEDs via GPIO + UART
Connectivity Backbone:
- All EFR32 boards on Thread protocol
- MG21 bridges Thread → Raspberry Pi 4
Embedded PC Layer (Raspberry Pi 4):
- Runs Customer Classification AI
- LCD Touch (iPad) user interface
- NLP Transformer for text understanding
Backend & Admin Management:
- Raspberry Pi ↔ Backend over LAN
- Pipeline: MQTT/Redis ↔ Flask ↔ WebSocket ↔ Frontend
- PostgreSQL + Admin Website