Research Paper

Core Innovation

This paper introduces DevPiolt, a large language model-based recommendation system fine-tuned with IoT domain knowledge and user preferences. It incorporates continual pre-training, multi-task fine-tuning, and a confidence-based exposure control mechanism to improve recommendation quality and avoid negative user experiences, outperforming existing baselines significantly.

Why It Matters

IoT users face challenges with complex device operations and diverse preferences, leading to poor recommendation relevance and user frustration. DevPiolt improves personalized operation suggestions, increasing user engagement and device utilization. This scalable solution enhances customer experience and drives business growth in smart home ecosystems.

Potential Customers & Pain Points

• Smart home platform providers – Need to increase user engagement and device utilization
• IoT device manufacturers – Need to improve customer satisfaction and reduce churn
• Mobile app developers for IoT – Need to deliver relevant personalized content
• Enterprises managing IoT ecosystems – Need to optimize device operation efficiency.

Business Model

Subscription and licensing model targeting smart home platform providers and IoT device manufacturers, with potential revenue from SaaS-based recommendation APIs and integration services.

Competitive Landscape

• Google Nest
• Amazon Alexa
• Samsung SmartThings
• Apple HomeKit

Implementation Challenges

• Integration complexity with diverse IoT devices and platforms
• User privacy and data security concerns
• Maintaining recommendation quality across heterogeneous user preferences

Validation Strategy

• Deploy in Xiaomi Home app with 255
• 000 users for real-world feedback
• Measure key metrics like device coverage and acceptance rates
• Conduct A/B testing against existing recommendation systems
• Collect user satisfaction surveys and iterate model improvements