WELLPET Transforms Smart Collar Telemetry into Safe, Real-Time Pet Health Insights with Amazon Bedrock
WellPet.ai was preparing for commercial launch, and their platform had to be able to scale quickly and reliably from day one.
They were already on AWS, but needed an experienced team to harden and stress test their architecture, and at the same time optimize the costs of running the solution.
They approached allOps because they know how deep we venture into every aspect of cloud, while keeping the underlying technology choice as simple as possible.
Project basics
- Project Production optimization of an AI-powered smart-collar behavioral insight and recommendation platform using Amazon Bedrock
- Industry Pet technology, connected devices, and animal health
Client
WellPet.ai develops a smart dog collar and companion app that helps owners understand their dog’s behavior, emotional state, and wellbeing. The collar captures behavioral and vocalization signals and translates them into clear, actionable insights.
Data collected from more than 100 dogs validated the underlying behavioral approach, and WellPet had more than 12,000 registered platform users.
It’s a premium canine health and behavior platform, built on a scientifically grounded approach developed in collaboration with veterinary behavior experts.
allOps did exactly that: pragmatic, no unnecessary complexity, delivered as agreed. We go to market with far more confidence in our platform.
We also appreciated that allOps optimized for our actual needs instead of adding technology for its own sake. Combined with their help around AWS programs, that makes them a partner we plan to work with again as we scale after launch.
Challenge
WellPet was preparing for commercial launch, and their platform had to be able to scale quickly and reliably from day one.
Their solution converts high-frequency smart-collar telemetry into timely, understandable, and safe behavioral insights for dog owners through large language models.
Behavioral and GPS parameters get transmitted approximately every five seconds. Bark detections are event-driven.
This means there are strict latency and scaling requirements across ingestion, processing, and recommendation generation.
They were already building on AWS with a serverless architecture and Amazon Bedrock as their Generative AI layer, but:
- They wanted an experienced partner to review and harden the platform before going to market
- They wanted the best possible balance between recommendation relevance, factual consistency, latency, safety, and inference cost when using LLMs
The customer had experimented with several foundation-model families, but couldn’t find the best fit across these vital areas.
Prompt behavior also varied between models, making manual prompt adjustment slow and increasing the effort required to validate pet-health-related recommendations before release.
So, the key challenge was not to simply select and call a valid LLM.
They required a repeatable method for:
- Selecting the appropriate foundation model
- Optimizing prompts against representative dog-behavior scenarios
- Integrating the chosen models with real-time structured telemetry
On top of that, the AI-generated content related to pet health has to remain informational and safe, so it passes through strict review and gating before reaching users.
Our Solution
Over an eight-week engagement from May through June 2026, allOps did essential architectural, safeguarding, and optimization work that enabled airtight AI-powered infrastructure for launch.
We production-hardened WellPets’s existing AWS serverless platform in the Europe (Frankfurt) eu-central-1 AWS Region and established Amazon Bedrock as the managed Generative AI inference layer.
Here’s more details about the specific seemingly smaller solutions that snowballed to a longlasting one:
- Collar devices perform initial signal detection before transmitting telemetry through AWS IoT Core. Amazon Kinesis Data Streams separated continuous behavioral and location telemetry from event-driven bark data
- AWS Lambda functions processes the streams and assembles a bounded context for each recommendation using the dog’s profile, recent sensor events, personalized behavioral baseline, and relevant recommendation history
- Amazon DynamoDB, Amazon S3, and Amazon ElastiCache supply durable and low-latency operational context
- Amazon API Gateway and AWS AppSync deliver generated insights to the companion application
- allOps replaced ad hoc model comparison with Amazon Bedrock Evaluations using a customer-reviewed golden-response dataset. Candidate models were assessed for correctness, recommendation relevance, coherence, safe handling of health-related signals, p95 latency, and normalized inference cost
- The evaluation selected Anthropic Claude Sonnet and Claude Opus model families through Amazon Bedrock: Sonnet for routine, high-volume recommendation generation and Opus for complex, multi-signal explanations requiring deeper reasoning
- Through Amazon Bedrock Prompt Management and Amazon Bedrock Advanced Prompt Optimization, we versioned and improved the prompts. The optimization process used representative inputs, expected responses, and defined evaluation criteria to compare original and optimized prompt variants before production promotion
- Amazon CloudWatch and AWS X-Ray measure latency, errors, model usage, guardrail actions, and end-to-end processing behavior
- AWS SAM and AWS CloudFormation provide repeatable infrastructure deployment
We intentionally did not introduce RAG because the authoritative input was current, structured, per-dog telemetry retrieved deterministically from operational data stores.
Foundation-model fine-tuning was also unnecessary because the required behavior and output structure met acceptance criteria through evaluated prompt templates.
Amazon Bedrock Guardrails now evaluate all user-facing generated recommendations. Deterministic application rules block diagnostic or medication claims, display explicit informational-use messaging, and direct owners to a veterinarian when they detect high-risk anomalies.
Results
WellPet now has a production-ready, safe, and scalable Generative AI foundation on AWS. All health-related content goes through safety gating.
Most of all, they have a measurable and repeatable process for selecting models, improving prompts, and controlling inference expenditure.
The architecture allOps helped harden and validate is what enables further growth and scaling.
The solution was released to the WellPet production environment and accepted on June 26 2026.
- 37% reduction in normalized generative AI inference cost
- Recommendation relevance increased from 80% to 94%
- 99.9% availability during the production acceptance period
- Amazon Bedrock Guardrails cover 100% of user-facing recommendations
About allOps Solutions
allOps is an AWS Advanced Tier Services Partner founded by two AWS Heroes – a distinction held by around 255 people worldwide – and the only AWS Authorized Training Partner in Bosnia and Herzegovina.
allOps designs, builds and operates secure, scalable AWS for companies from startups to global enterprises, turning AWS funding into outcomes customers can see.
Apply for an AWS-funded foundation sprint and push your GenAI PoC to production today.