Qualcomm’s Dragonwing IQ-9075 EVK demonstrates edge-based GenAI using two Qualcomm AI runtimes: the Genie software library for LLM inference with RAG, and Whisper ASR for speech-to-text conversion. Three scenarios were implemented: Predictive Maintenance, Factory Management, and Demo Mode. The solution uses three AI models: Meta Llama 3.2 Instruct 3B for LLM, BAAI BGE-large for RAG embeddings, and OpenAI Whisper-small for STT. All models are optimized for Hexagon Tensor Processor (HTP) accelerators, with workloads distributed across the two HTP cores of the Dragonwing IQ-9075 EVK, enabling simultaneous ASR and LLM with RAG on a single embedded device. The Predictive Maintenance use case ingests static documentation—manuals, maintenance records, and inspection reports—to surface relevant information for technicians, enabling proactive issue identification and optimized service scheduling. The solution can be deployed directly on devices or local gateways, ensuring maintenance insights remain available in limited-connectivity environments. The RAG architecture centers on ChromaDB: technical documentation is encoded into dense vector representations using the BGE-large embedding model, semantic search retrieves relevant maintenance instructions, and Llama 3.2 3B synthesizes the final response. This enables precise semantic retrieval rather than keyword matching. Llama 3.2 3B runs directly on the Dragonwing IQ-9075 Hexagon