AWS Builder Center
Agents for Humans: PetWhisperer with the Strands Agents SDK

Agents for Humans: PetWhisperer with the Strands Agents SDK

AWS AgentCore Architecture By the PetWhisperer Engineering Team Category: AI & Machine Learning, Architecture, Serverless, IoT

🌟 The Vision: Why "Quiet" Background AI Matters for Humans & Pets Modern smart technology is demanding. Every device we bring into our homes demands attention: sending notifications, asking for confirmations, and pulling us away from the present moment. For the 85+ million households caring for companion animals, pet-care tech has suffered from the same flawβ€”requiring constant manual logging, video scrubbing, and screen interaction.
We asked: What if autonomous AI could work silently on behalf of humans and their pets?
An intelligent companion that:
Listens continuously to environmental acoustics and smart collar biosensors in the background (running at >95% quiet mode).
De-escalates acute animal anxiety within seconds using dynamic bio-acoustic Solfeggio sound waves.
Automatically drafts structured veterinary SOAP records and syncs with clinical EHRs.
Only surfaces to the human when a high-stakes decision is truly required, following a strict, 60-word voice-actionable protocol (vb_call).
This is PetWhispererβ€”a cross-species AI guardian powered by the Strands Agents SDK, AWS AgentCore patterns, and Gemini Multimodal Vision.
πŸ—οΈ System Architecture: The 5-Tier Agentic Mesh code Code β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 1. MULTIMODAL INGESTION β”‚ β”‚ Smart Collar BLE 5.3 PPG (HR/HRV) β€’ 25Hz Audio FFT β€’ DogFACS Video β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 2. EDGE COMPILER & LOCAL SIMD INFERENCE β”‚ β”‚ ONNX INT8 / WASM SIMD Runtime (<10ms air-gapped response) β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 3. STRANDS MULTI-TRACK AGENT MESH β”‚ β”‚ Track 1: Everyday Home β”‚ Track 2: Clinical SOAP β”‚ Track 3: Neighbor β”‚ β”‚ Guardian Agent β”‚ Veterinary Agent β”‚ Mesh Relay β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 4. CLINICAL EHR & INTEROPERABILITY GATEWAY β”‚ β”‚ Bidirectional HL7 FHIR v4.0.1 (Idexx Cornerstone / Covetrus / ezyVet)β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β”‚ β–Ό β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ 5. ENTERPRISE AWS AGENTCORE FLEET MANAGEMENT β”‚ β”‚ Zero-Trust RBAC β€’ AWS Bedrock Agents β€’ EventBridge β€’ DynamoDB β”‚ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ ⚑ How We Built It Using AWS & The Strands Agents SDK
  1. Multi-Track Agent Mesh with AWS AgentCore Patterns PetWhisperer splits complex real-world workflows into three decoupled agent tracks using the Strands Agents SDK:
Track 1: Everyday Home Agents (Silent Sensory Regulation)
Subscribes to ambient acoustic decibels and smart collar PPG telemetry.
When an ambient trigger (thunderstorm or appliance hum) is detected, the agent autonomously triggers a 432Hz/528Hz Solfeggio soothing audio wave via the Web Audio API without human interruption.
Automatically manages pet pantry replenishment and reorders food when supply thresholds drop.
Track 2: Professional Clinical Agents (Veterinary SOAP Scribing)
Transcribes live veterinary consultations and maps animal micro-gestures to standardized DogFACS (Canine), CatFACS (Feline), and EquiFACS (Equine) action units.
Compiles structured SOAP (Subjective, Objective, Assessment, Plan) progress notes and pushes bidirectional transaction bundles to HL7 FHIR v4.0.1 clinical gateways.
Track 3: Good Neighbor Agents (Community Mutual Aid)
Operates a decentralized lost-pet search grid, mapping sighting vectors and coordinating emergency shelter foster relays during extreme weather events.
  1. Edge-Ready Compilers for Zero-Cloud Latency To ensure pet safety during internet outages, we compiled our bio-acoustic FFT audio models and facial action unit classifiers into lightweight ONNX INT8, TensorRT, and WebAssembly (WASM SIMD) micro-binaries.
Inference Latency: Sub-10ms execution on local hardware (Raspberry Pi 5, Apple HomePod, Coral Edge TPU, and smart pet camera SoCs).
Zero-Cloud Privacy: Full air-gapped operation with 0.0 KB outbound traffic when privacy mode is engaged.
  1. The vb_call Human Escalation Guardrail One of our proudest architectural decisions is the strict escalation calibration:
Minor routine actions (sound wave soothing, light temperature adjustment, routine telemetry logging) run 100% autonomously.
High-stakes decisions (prescription dosage questions, emergency veterinary trauma, expenses > $50) generate a concise 60-word voice card:
The Decision (first 10 words)
The Situation
The Numbered Choices (Option 1 vs. Option 2)
The Stakes
πŸ’‘ Lessons Learned & What's Next Autonomous Agents Must Be Quiet: High-trust agentic systems earn user confidence through reliable background execution and transparent telemetry logs, not constant push alerts.
Cross-Species Ethology is Multimodal: Acoustic frequency analysis combined with micro-facial action coding (like sclera whale-eye exposure or ear pinna flattening) detects autonomic stress long before behavioral escalation occurs.
What's Next: Direct Smart Collar Firmware Integration: Ingesting raw BLE 5.3 ISO streams directly from commercial smart collars (Fi, Invoxia, PetPace).
AWS Bedrock Multi-Agent Swarms: Scaling our collaborative partner memory and community lost-pet search grids using Amazon Bedrock Agentic workflows.
Built with passion for the AWS & AI Studio Hackathons. Check out the live application and test the multi-species ethology pipeline today!
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article