AWS details IskayPet's corporate knowledge assistant built on Amazon Bedrock Managed Knowledge Bases

What AWS announced

AWS published a blog post describing how IskayPet, a pet care retailer operating under the Kiwoko, Tiendanimal, Kivet, Clinicanimal and Animalis brands across Spain, Portugal and France, built an internal knowledge assistant using Amazon Bedrock Managed Knowledge Bases.

According to AWS, IskayPet runs more than 330 stores and employs more than 4,500 people serving 7 million customers. AWS says the company's operational knowledge, covering return policies, veterinary protocols, onboarding steps and procurement procedures, was spread across hundreds of Microsoft SharePoint documents in four languages, and that finding the right document took employees significant time.

The architecture AWS describes

AWS states that IskayPet selected Amazon Bedrock Managed Knowledge Bases specifically to avoid operating vector databases, embedding pipelines and ingestion jobs directly. AWS lists several capabilities it says drove the decision:

• A native SharePoint connector that AWS says ingests documents directly from IskayPet's existing environment, avoiding a data migration, and processes content from images embedded in documents when advanced indexing is enabled.
• Built-in hybrid search, combining semantic and keyword (BM25) search, which AWS says recovers exact terms such as product codes that pure semantic search can miss.
• Managed reranking of retrieved passages, which AWS says is included at no additional charge.
• Document-level access control, which AWS says enforces SharePoint's own permissions at retrieval time so a query returns only passages from documents the requesting employee is already allowed to read.
• Model flexibility, letting IskayPet change the underlying model later without rewriting infrastructure, according to AWS.
• Built-in RAG evaluation tools, which AWS says the team used to measure quality.

AWS describes the resulting flow as follows: an employee submits a question through a chat interface; the application calls the knowledge base for the relevant assistant, passing the employee's identity so permissions apply to retrieval; the knowledge base runs hybrid search, reranks results, and returns passages with source references; a foundation model then generates an answer from those passages, with a prompt that AWS says requires source citation and instructs the model to state when an answer isn't in the retrieved material.

Scope and development

AWS says the assistant answers questions across retail operations, veterinary care, people and onboarding, and procurement, in Spanish, Portuguese, French and English, and across IskayPet's three countries of operation. AWS states the goal was for the system to route each question to the correct body of knowledge without cross-contamination and to scale to new areas or countries without requiring a re-architecture.

AWS also states that IskayPet's team built the complete solution, including the store-facing frontend, the AWS architecture, and the deployment, using Kiro with Claude Haiku 4.5.

Source: AWS Machine Learning Blog, "How IskayPet unified corporate knowledge with Amazon Bedrock Managed Knowledge Bases," published October 22, 2026.