📲🎮𖣘🆒 Can we make classic cooling systems smarter using edge AI and physical simulations? During the severe heatwaves early this summer, I noticed how ubiquitous ceiling fans remain in historic markets, public buildings, and homes. Despite being cost-effective, they lack smart airflow optimization compared to modern AC systems. To bridge this gap, I built a complete, end-to-end proof-of-concept system powered by edge AI and local vision-language models (VLMs). Here is a look at the engineering stack behind the project: 📐 Custom Differential Bevel Gear Mechanism: Designed a custom 2-fan setup moving 180° vertically and 360° horizontally to enable dynamic airflow control. Built on a scaled physical setup with an IKEA LACK table base, mirrored perfectly as a 3D digital twin in Fusion 360. 📡 2D LiDAR to 3D Obstacle Mapping: Integrated an RPLIDAR A1M8-R6 to map real-time indoor obstacles (walls, open doors, furniture) and converted 2D scan points into a 3D obstacle map for dynamic boundary collision tracking. 🎮 Physics and Airflow Simulation in @unity: Utilized Unity as a real-time rendering and simulation engine to trace air displacement routes (trails), capturing frame-by-frame airflow dynamics across different mechanical fan configurations. 🧠 On-Device Vision-Language Models (VLMs): Deployed local open-weight VLMs (Qwen2.5-VL-7B-Instruct and Qwen3-VL-8B-Instruct) to evaluate visual simulation frames and output optimal fan configurations for maximum cooling efficiency, completely offline. 💻 Edge Computing on @Qualcomm Dragonwing IQ-9075 EVK: Ran the complete stack on a single edge platform by distributing workloads across hardware blocks: 🔵 Kryo CPU: LiDAR calculations and processing 🔵 Adreno GPU: Unity scene and airflow simulation rendering 🔵 Hexagon NPU: Local VLM inference via Qualcomm Docker LLM/VLM container 📲 Embedded Control and Custom PCB Design: 🟢 Built a custom 4-layer PCB around an @arduino Nano 33 BLE to manage stepper drivers, DC motors, encoders, optical speed modules, and lighting. 🟢 Developed an Android app with embedded Flask middleware, keeping the user in the loop for safe manual overrides and VLM prompt interactions. By pairing lightweight spatial hardware with Edge AI, we can unlock significant energy-wise and coverage-wise cooling performance in everyday appliances without relying on cloud infrastructure. Check out the full technical write-up and tutorial on @Hacksterio 👇 hackster.io/kutluhan-aktar/a… @Qualcomm_Dev #EdgeAI #EmbeddedSystems #Qualcomm #Dragonwing #Robotics #IoT #ComputerVision #VLMs #Unity #Unity3D #Arduino #PCBDesign #HardwareEngineering #ArtificialIntelligence