We’re developing a smart HVAC ecosystem that will eventually use and embed AI directly into the Hub, enabling it to learn from sensor patterns and user behavior. This project focuses on researching lightweight, local-first AI/ML frameworks suitable for use in embedded devices (e.g., ESP32, Raspberry Pi, or other microcontrollers). Students will identify, compare, and evaluate existing open-source or commercial frameworks that could be used for future on-device learning, pattern detection, and predictive control in a privacy-first HVAC environment. The results will help us choose the right tools for smart scheduling, predictive heating/cooling, and future SaaS or licensing features — all while respecting device constraints and privacy.

We will provide students with:

• A basic outline of the system
• Controller architecture (ESP32 class devices)
• Sample use cases (e.g., optimizing temp/humidity scheduling)
• Constraints (local-first, no cloud dependence, 4MB flash, Wi-Fi only)

Deliverables:

• Evaluation matrix of 5–10 AI tools (e.g., Edge Impulse, TensorFlow Lite, TinyML, SensiML)
• Comparison chart: memory footprint, processing needs, offline capability, licensing
• Short report outlining suitability for HVAC use cases (e.g., temperature prediction, occupancy patterns)
• GitHub examples or whitepapers where applicable

Student Learning Outcomes:

• Understanding edge-AI tradeoffs (local vs cloud, energy vs capability)
• Researching AI frameworks with embedded/IoT constraints
• Translating product needs into technical tool evaluation