• Record audio/video tasks that help train next-generation AI systems. • Review requirements, then continue to partner onboarding. • You will participate in audio/video recording tasks that may involve speaking, labeling, reviewing, or following on-screen instructions. • Your submissions support the development and training of next-generation AI systems. • We are collecting high-quality human data to improve how AI understands speech, behavior, communication, and real-world interactions. • Follow task instructions carefully and record in a clear, accurate, and suitable environment.

• Record audio/video tasks that help train next-generation AI systems. • Review requirements, then continue to partner onboarding. • Participate in audio/video recording tasks involving speaking, labeling, reviewing, or following on-screen instructions. • Support development and training of next-generation AI systems. • Collect high-quality human data to improve AI understanding of speech and communication.

• Interact with your assigned AI model on designated health and wellness topics. • Use your personal AI chatbot account and existing conversation history to evaluate personalized response quality. • Assess how well the system adapts to your inputs and provides accurate, relevant responses. • Address feedback and implement revisions to support high quality submissions. • Complete a structured evaluation form based on your experience.

• Lead SLM candidate evaluation and selection: assess Small Language Model options for edge deployment against hardware constraints, inference latency requirements, domain restriction feasibility, and licensing. • Produce a technology assessment with explicit trade-off rationale and a recommended approach. • Design the domain restriction and guardrails architecture: define how the SLM is constrained to a known operational scope, how out-of-domain responses are prevented, and how the system enforces retrieval-first, non-authoritative behavior appropriate for a safety-adjacent environment. • Design the capability framework that structures how the system responds to operator queries — how capabilities are scoped and isolated, how the framework supports incremental addition of new interaction types over time, and what the prototype will implement. • Design the retrieval-augmented inference pipeline: define how the SLM retrieves context from a local knowledge store at inference time, including retrieval strategy, context injection approach, and latency budget appropriate for the edge environment. • Evaluate candidate cloud services for knowledge retrieval, model governance, and fleet-level model lifecycle management including over-the-air model distribution to edge devices. • Produce architecture recommendations aligned to client enterprise standards; all service selections are subject to client review and approval. • Define the offboard ML lifecycle: how models are evaluated, adapted through prompting and retrieval augmentation, versioned, governed, and distributed at scale. • Fine-tuning or custom model training is not a default commitment in this phase — adaptation approach will be determined based on discovery findings. • Collaborate with the Edge ML / Embedded Engineer on hardware constraint inputs that shape SLM selection and inference pipeline design, ensuring architecture recommendations are grounded in confirmed runtime feasibility. • Collaborate with the AWS Solutions Architect on candidate cloud service architecture for model governance, knowledge retrieval, and the model update pipeline, ensuring the cloud-side AI architecture aligns with the broader platform. • Document safety design principles and operational boundaries — authority separation, bounded AI behavior, explainability approach, and human-in-the-loop considerations — as architecture artifacts for client engineering and compliance review. • Produce all architecture recommendations as Architecture Decision Records (ADRs) with explicit trade-off rationale. • Clearly distinguish confirmed decisions from those that remain conditional on hardware specifications or interface access not yet confirmed.